Arulkumar03/Wheat_HEAD_Detection_Counting_ComputerVision_Model
0
1# Detectron2 Model Zoo and Baselines2 3## Introduction4 5This file documents a large collection of baselines trained6with detectron2 in Sep-Oct, 2019.7All numbers were obtained on [Big Basin](https://engineering.fb.com/data-center-engineering/introducing-big-basin-our-next-generation-ai-hardware/)8servers with 8 NVIDIA V100 GPUs & NVLink. The speed numbers are periodically updated with latest PyTorch/CUDA/cuDNN versions.9You can access these models from code using [detectron2.model_zoo](https://detectron2.readthedocs.io/modules/model_zoo.html) APIs.10 11In addition to these official baseline models, you can find more models in [projects/](projects/).12 13#### How to Read the Tables14* The "Name" column contains a link to the config file. Models can be reproduced using `tools/train_net.py` with the corresponding yaml config file,15 or `tools/lazyconfig_train_net.py` for python config files.16* Training speed is averaged across the entire training.17 We keep updating the speed with latest version of detectron2/pytorch/etc.,18 so they might be different from the `metrics` file.19 Training speed for multi-machine jobs is not provided.20* Inference speed is measured by `tools/train_net.py --eval-only`, or [inference_on_dataset()](https://detectron2.readthedocs.io/modules/evaluation.html#detectron2.evaluation.inference_on_dataset),21 with batch size 1 in detectron2 directly.22 Measuring it with custom code may introduce other overhead.23 Actual deployment in production should in general be faster than the given inference24 speed due to more optimizations.25* The *model id* column is provided for ease of reference.26 To check downloaded file integrity, any model on this page contains its md5 prefix in its file name.27* Training curves and other statistics can be found in `metrics` for each model.28 29#### Common Settings for COCO Models30* All COCO models were trained on `train2017` and evaluated on `val2017`.31* The default settings are __not directly comparable__ with Detectron's standard settings.32 For example, our default training data augmentation uses scale jittering in addition to horizontal flipping.33 34 To make fair comparisons with Detectron's settings, see35 [Detectron1-Comparisons](configs/Detectron1-Comparisons/) for accuracy comparison,36 and [benchmarks](https://detectron2.readthedocs.io/notes/benchmarks.html)37 for speed comparison.38* For Faster/Mask R-CNN, we provide baselines based on __3 different backbone combinations__:39 * __FPN__: Use a ResNet+FPN backbone with standard conv and FC heads for mask and box prediction,40 respectively. It obtains the best41 speed/accuracy tradeoff, but the other two are still useful for research.42 * __C4__: Use a ResNet conv4 backbone with conv5 head. The original baseline in the Faster R-CNN paper.43 * __DC5__ (Dilated-C5): Use a ResNet conv5 backbone with dilations in conv5, and standard conv and FC heads44 for mask and box prediction, respectively.45 This is used by the Deformable ConvNet paper.46* Most models are trained with the 3x schedule (~37 COCO epochs).47 Although 1x models are heavily under-trained, we provide some ResNet-50 models with the 1x (~12 COCO epochs)48 training schedule for comparison when doing quick research iteration.49 50#### ImageNet Pretrained Models51 52It's common to initialize from backbone models pre-trained on ImageNet classification tasks. The following backbone models are available:53 54* [R-50.pkl](https://dl.fbaipublicfiles.com/detectron2/ImageNetPretrained/MSRA/R-50.pkl): converted copy of [MSRA's original ResNet-50](https://github.com/KaimingHe/deep-residual-networks) model.55* [R-101.pkl](https://dl.fbaipublicfiles.com/detectron2/ImageNetPretrained/MSRA/R-101.pkl): converted copy of [MSRA's original ResNet-101](https://github.com/KaimingHe/deep-residual-networks) model.56* [X-101-32x8d.pkl](https://dl.fbaipublicfiles.com/detectron2/ImageNetPretrained/FAIR/X-101-32x8d.pkl): ResNeXt-101-32x8d model trained with Caffe2 at FB.57* [R-50.pkl (torchvision)](https://dl.fbaipublicfiles.com/detectron2/ImageNetPretrained/torchvision/R-50.pkl): converted copy of [torchvision's ResNet-50](https://pytorch.org/docs/stable/torchvision/models.html#torchvision.models.resnet50) model.58 More details can be found in [the conversion script](tools/convert-torchvision-to-d2.py).59 60Note that the above models have __different__ format from those provided in Detectron: we do not fuse BatchNorm into an affine layer.61Pretrained models in Detectron's format can still be used. For example:62* [X-152-32x8d-IN5k.pkl](https://dl.fbaipublicfiles.com/detectron/ImageNetPretrained/25093814/X-152-32x8d-IN5k.pkl):63 ResNeXt-152-32x8d model trained on ImageNet-5k with Caffe2 at FB (see ResNeXt paper for details on ImageNet-5k).64* [R-50-GN.pkl](https://dl.fbaipublicfiles.com/detectron/ImageNetPretrained/47261647/R-50-GN.pkl):65 ResNet-50 with Group Normalization.66* [R-101-GN.pkl](https://dl.fbaipublicfiles.com/detectron/ImageNetPretrained/47592356/R-101-GN.pkl):67 ResNet-101 with Group Normalization.68 69These models require slightly different settings regarding normalization and architecture. See the model zoo configs for reference.70 71#### License72 73All models available for download through this document are licensed under the74[Creative Commons Attribution-ShareAlike 3.0 license](https://creativecommons.org/licenses/by-sa/3.0/).75 76### COCO Object Detection Baselines77 78#### Faster R-CNN:79<!--80(fb only) To update the table in vim:811. Remove the old table: d}822. Copy the below command to the place of the table833. :.!bash84 85./gen_html_table.py --config 'COCO-Detection/faster*50*'{1x,3x}'*' 'COCO-Detection/faster*101*' --name R50-C4 R50-DC5 R50-FPN R50-C4 R50-DC5 R50-FPN R101-C4 R101-DC5 R101-FPN X101-FPN --fields lr_sched train_speed inference_speed mem box_AP86-->87 88 89<table><tbody>90<!-- START TABLE -->91<!-- TABLE HEADER -->92<th valign="bottom">Name</th>93<th valign="bottom">lr<br/>sched</th>94<th valign="bottom">train<br/>time<br/>(s/iter)</th>95<th valign="bottom">inference<br/>time<br/>(s/im)</th>96<th valign="bottom">train<br/>mem<br/>(GB)</th>97<th valign="bottom">box<br/>AP</th>98<th valign="bottom">model id</th>99<th valign="bottom">download</th>100<!-- TABLE BODY -->101<!-- ROW: faster_rcnn_R_50_C4_1x -->102 <tr><td align="left"><a href="configs/COCO-Detection/faster_rcnn_R_50_C4_1x.yaml">R50-C4</a></td>103<td align="center">1x</td>104<td align="center">0.551</td>105<td align="center">0.102</td>106<td align="center">4.8</td>107<td align="center">35.7</td>108<td align="center">137257644</td>109<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_C4_1x/137257644/model_final_721ade.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_C4_1x/137257644/metrics.json">metrics</a></td>110</tr>111<!-- ROW: faster_rcnn_R_50_DC5_1x -->112 <tr><td align="left"><a href="configs/COCO-Detection/faster_rcnn_R_50_DC5_1x.yaml">R50-DC5</a></td>113<td align="center">1x</td>114<td align="center">0.380</td>115<td align="center">0.068</td>116<td align="center">5.0</td>117<td align="center">37.3</td>118<td align="center">137847829</td>119<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_DC5_1x/137847829/model_final_51d356.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_DC5_1x/137847829/metrics.json">metrics</a></td>120</tr>121<!-- ROW: faster_rcnn_R_50_FPN_1x -->122 <tr><td align="left"><a href="configs/COCO-Detection/faster_rcnn_R_50_FPN_1x.yaml">R50-FPN</a></td>123<td align="center">1x</td>124<td align="center">0.210</td>125<td align="center">0.038</td>126<td align="center">3.0</td>127<td align="center">37.9</td>128<td align="center">137257794</td>129<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_FPN_1x/137257794/model_final_b275ba.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_FPN_1x/137257794/metrics.json">metrics</a></td>130</tr>131<!-- ROW: faster_rcnn_R_50_C4_3x -->132 <tr><td align="left"><a href="configs/COCO-Detection/faster_rcnn_R_50_C4_3x.yaml">R50-C4</a></td>133<td align="center">3x</td>134<td align="center">0.543</td>135<td align="center">0.104</td>136<td align="center">4.8</td>137<td align="center">38.4</td>138<td align="center">137849393</td>139<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_C4_3x/137849393/model_final_f97cb7.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_C4_3x/137849393/metrics.json">metrics</a></td>140</tr>141<!-- ROW: faster_rcnn_R_50_DC5_3x -->142 <tr><td align="left"><a href="configs/COCO-Detection/faster_rcnn_R_50_DC5_3x.yaml">R50-DC5</a></td>143<td align="center">3x</td>144<td align="center">0.378</td>145<td align="center">0.070</td>146<td align="center">5.0</td>147<td align="center">39.0</td>148<td align="center">137849425</td>149<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_DC5_3x/137849425/model_final_68d202.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_DC5_3x/137849425/metrics.json">metrics</a></td>150</tr>151<!-- ROW: faster_rcnn_R_50_FPN_3x -->152 <tr><td align="left"><a href="configs/COCO-Detection/faster_rcnn_R_50_FPN_3x.yaml">R50-FPN</a></td>153<td align="center">3x</td>154<td align="center">0.209</td>155<td align="center">0.038</td>156<td align="center">3.0</td>157<td align="center">40.2</td>158<td align="center">137849458</td>159<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_FPN_3x/137849458/model_final_280758.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_50_FPN_3x/137849458/metrics.json">metrics</a></td>160</tr>161<!-- ROW: faster_rcnn_R_101_C4_3x -->162 <tr><td align="left"><a href="configs/COCO-Detection/faster_rcnn_R_101_C4_3x.yaml">R101-C4</a></td>163<td align="center">3x</td>164<td align="center">0.619</td>165<td align="center">0.139</td>166<td align="center">5.9</td>167<td align="center">41.1</td>168<td align="center">138204752</td>169<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_101_C4_3x/138204752/model_final_298dad.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_101_C4_3x/138204752/metrics.json">metrics</a></td>170</tr>171<!-- ROW: faster_rcnn_R_101_DC5_3x -->172 <tr><td align="left"><a href="configs/COCO-Detection/faster_rcnn_R_101_DC5_3x.yaml">R101-DC5</a></td>173<td align="center">3x</td>174<td align="center">0.452</td>175<td align="center">0.086</td>176<td align="center">6.1</td>177<td align="center">40.6</td>178<td align="center">138204841</td>179<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_101_DC5_3x/138204841/model_final_3e0943.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_101_DC5_3x/138204841/metrics.json">metrics</a></td>180</tr>181<!-- ROW: faster_rcnn_R_101_FPN_3x -->182 <tr><td align="left"><a href="configs/COCO-Detection/faster_rcnn_R_101_FPN_3x.yaml">R101-FPN</a></td>183<td align="center">3x</td>184<td align="center">0.286</td>185<td align="center">0.051</td>186<td align="center">4.1</td>187<td align="center">42.0</td>188<td align="center">137851257</td>189<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_101_FPN_3x/137851257/model_final_f6e8b1.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_R_101_FPN_3x/137851257/metrics.json">metrics</a></td>190</tr>191<!-- ROW: faster_rcnn_X_101_32x8d_FPN_3x -->192 <tr><td align="left"><a href="configs/COCO-Detection/faster_rcnn_X_101_32x8d_FPN_3x.yaml">X101-FPN</a></td>193<td align="center">3x</td>194<td align="center">0.638</td>195<td align="center">0.098</td>196<td align="center">6.7</td>197<td align="center">43.0</td>198<td align="center">139173657</td>199<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_X_101_32x8d_FPN_3x/139173657/model_final_68b088.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/faster_rcnn_X_101_32x8d_FPN_3x/139173657/metrics.json">metrics</a></td>200</tr>201</tbody></table>202 203#### RetinaNet:204<!--205./gen_html_table.py --config 'COCO-Detection/retina*50*' 'COCO-Detection/retina*101*' --name R50 R50 R101 --fields lr_sched train_speed inference_speed mem box_AP206-->207 208<table><tbody>209<!-- START TABLE -->210<!-- TABLE HEADER -->211<th valign="bottom">Name</th>212<th valign="bottom">lr<br/>sched</th>213<th valign="bottom">train<br/>time<br/>(s/iter)</th>214<th valign="bottom">inference<br/>time<br/>(s/im)</th>215<th valign="bottom">train<br/>mem<br/>(GB)</th>216<th valign="bottom">box<br/>AP</th>217<th valign="bottom">model id</th>218<th valign="bottom">download</th>219<!-- TABLE BODY -->220<!-- ROW: retinanet_R_50_FPN_1x -->221 <tr><td align="left"><a href="configs/COCO-Detection/retinanet_R_50_FPN_1x.yaml">R50</a></td>222<td align="center">1x</td>223<td align="center">0.205</td>224<td align="center">0.041</td>225<td align="center">4.1</td>226<td align="center">37.4</td>227<td align="center">190397773</td>228<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/retinanet_R_50_FPN_1x/190397773/model_final_bfca0b.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/retinanet_R_50_FPN_1x/190397773/metrics.json">metrics</a></td>229</tr>230<!-- ROW: retinanet_R_50_FPN_3x -->231 <tr><td align="left"><a href="configs/COCO-Detection/retinanet_R_50_FPN_3x.yaml">R50</a></td>232<td align="center">3x</td>233<td align="center">0.205</td>234<td align="center">0.041</td>235<td align="center">4.1</td>236<td align="center">38.7</td>237<td align="center">190397829</td>238<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/retinanet_R_50_FPN_3x/190397829/model_final_5bd44e.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/retinanet_R_50_FPN_3x/190397829/metrics.json">metrics</a></td>239</tr>240<!-- ROW: retinanet_R_101_FPN_3x -->241 <tr><td align="left"><a href="configs/COCO-Detection/retinanet_R_101_FPN_3x.yaml">R101</a></td>242<td align="center">3x</td>243<td align="center">0.291</td>244<td align="center">0.054</td>245<td align="center">5.2</td>246<td align="center">40.4</td>247<td align="center">190397697</td>248<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/retinanet_R_101_FPN_3x/190397697/model_final_971ab9.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/retinanet_R_101_FPN_3x/190397697/metrics.json">metrics</a></td>249</tr>250</tbody></table>251 252 253#### RPN & Fast R-CNN:254<!--255./gen_html_table.py --config 'COCO-Detection/rpn*' 'COCO-Detection/fast_rcnn*' --name "RPN R50-C4" "RPN R50-FPN" "Fast R-CNN R50-FPN" --fields lr_sched train_speed inference_speed mem box_AP prop_AR256-->257 258<table><tbody>259<!-- START TABLE -->260<!-- TABLE HEADER -->261<th valign="bottom">Name</th>262<th valign="bottom">lr<br/>sched</th>263<th valign="bottom">train<br/>time<br/>(s/iter)</th>264<th valign="bottom">inference<br/>time<br/>(s/im)</th>265<th valign="bottom">train<br/>mem<br/>(GB)</th>266<th valign="bottom">box<br/>AP</th>267<th valign="bottom">prop.<br/>AR</th>268<th valign="bottom">model id</th>269<th valign="bottom">download</th>270<!-- TABLE BODY -->271<!-- ROW: rpn_R_50_C4_1x -->272 <tr><td align="left"><a href="configs/COCO-Detection/rpn_R_50_C4_1x.yaml">RPN R50-C4</a></td>273<td align="center">1x</td>274<td align="center">0.130</td>275<td align="center">0.034</td>276<td align="center">1.5</td>277<td align="center"></td>278<td align="center">51.6</td>279<td align="center">137258005</td>280<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/rpn_R_50_C4_1x/137258005/model_final_450694.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/rpn_R_50_C4_1x/137258005/metrics.json">metrics</a></td>281</tr>282<!-- ROW: rpn_R_50_FPN_1x -->283 <tr><td align="left"><a href="configs/COCO-Detection/rpn_R_50_FPN_1x.yaml">RPN R50-FPN</a></td>284<td align="center">1x</td>285<td align="center">0.186</td>286<td align="center">0.032</td>287<td align="center">2.7</td>288<td align="center"></td>289<td align="center">58.0</td>290<td align="center">137258492</td>291<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/rpn_R_50_FPN_1x/137258492/model_final_02ce48.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/rpn_R_50_FPN_1x/137258492/metrics.json">metrics</a></td>292</tr>293<!-- ROW: fast_rcnn_R_50_FPN_1x -->294 <tr><td align="left"><a href="configs/COCO-Detection/fast_rcnn_R_50_FPN_1x.yaml">Fast R-CNN R50-FPN</a></td>295<td align="center">1x</td>296<td align="center">0.140</td>297<td align="center">0.029</td>298<td align="center">2.6</td>299<td align="center">37.8</td>300<td align="center"></td>301<td align="center">137635226</td>302<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/fast_rcnn_R_50_FPN_1x/137635226/model_final_e5f7ce.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Detection/fast_rcnn_R_50_FPN_1x/137635226/metrics.json">metrics</a></td>303</tr>304</tbody></table>305 306### COCO Instance Segmentation Baselines with Mask R-CNN307<!--308./gen_html_table.py --config 'COCO-InstanceSegmentation/mask*50*'{1x,3x}'*' 'COCO-InstanceSegmentation/mask*101*' --name R50-C4 R50-DC5 R50-FPN R50-C4 R50-DC5 R50-FPN R101-C4 R101-DC5 R101-FPN X101-FPN --fields lr_sched train_speed inference_speed mem box_AP mask_AP309-->310 311 312 313<table><tbody>314<!-- START TABLE -->315<!-- TABLE HEADER -->316<th valign="bottom">Name</th>317<th valign="bottom">lr<br/>sched</th>318<th valign="bottom">train<br/>time<br/>(s/iter)</th>319<th valign="bottom">inference<br/>time<br/>(s/im)</th>320<th valign="bottom">train<br/>mem<br/>(GB)</th>321<th valign="bottom">box<br/>AP</th>322<th valign="bottom">mask<br/>AP</th>323<th valign="bottom">model id</th>324<th valign="bottom">download</th>325<!-- TABLE BODY -->326<!-- ROW: mask_rcnn_R_50_C4_1x -->327 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_R_50_C4_1x.yaml">R50-C4</a></td>328<td align="center">1x</td>329<td align="center">0.584</td>330<td align="center">0.110</td>331<td align="center">5.2</td>332<td align="center">36.8</td>333<td align="center">32.2</td>334<td align="center">137259246</td>335<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_C4_1x/137259246/model_final_9243eb.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_C4_1x/137259246/metrics.json">metrics</a></td>336</tr>337<!-- ROW: mask_rcnn_R_50_DC5_1x -->338 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_R_50_DC5_1x.yaml">R50-DC5</a></td>339<td align="center">1x</td>340<td align="center">0.471</td>341<td align="center">0.076</td>342<td align="center">6.5</td>343<td align="center">38.3</td>344<td align="center">34.2</td>345<td align="center">137260150</td>346<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_DC5_1x/137260150/model_final_4f86c3.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_DC5_1x/137260150/metrics.json">metrics</a></td>347</tr>348<!-- ROW: mask_rcnn_R_50_FPN_1x -->349 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_1x.yaml">R50-FPN</a></td>350<td align="center">1x</td>351<td align="center">0.261</td>352<td align="center">0.043</td>353<td align="center">3.4</td>354<td align="center">38.6</td>355<td align="center">35.2</td>356<td align="center">137260431</td>357<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_1x/137260431/model_final_a54504.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_1x/137260431/metrics.json">metrics</a></td>358</tr>359<!-- ROW: mask_rcnn_R_50_C4_3x -->360 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_R_50_C4_3x.yaml">R50-C4</a></td>361<td align="center">3x</td>362<td align="center">0.575</td>363<td align="center">0.111</td>364<td align="center">5.2</td>365<td align="center">39.8</td>366<td align="center">34.4</td>367<td align="center">137849525</td>368<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_C4_3x/137849525/model_final_4ce675.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_C4_3x/137849525/metrics.json">metrics</a></td>369</tr>370<!-- ROW: mask_rcnn_R_50_DC5_3x -->371 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_R_50_DC5_3x.yaml">R50-DC5</a></td>372<td align="center">3x</td>373<td align="center">0.470</td>374<td align="center">0.076</td>375<td align="center">6.5</td>376<td align="center">40.0</td>377<td align="center">35.9</td>378<td align="center">137849551</td>379<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_DC5_3x/137849551/model_final_84107b.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_DC5_3x/137849551/metrics.json">metrics</a></td>380</tr>381<!-- ROW: mask_rcnn_R_50_FPN_3x -->382 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml">R50-FPN</a></td>383<td align="center">3x</td>384<td align="center">0.261</td>385<td align="center">0.043</td>386<td align="center">3.4</td>387<td align="center">41.0</td>388<td align="center">37.2</td>389<td align="center">137849600</td>390<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x/137849600/model_final_f10217.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x/137849600/metrics.json">metrics</a></td>391</tr>392<!-- ROW: mask_rcnn_R_101_C4_3x -->393 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_R_101_C4_3x.yaml">R101-C4</a></td>394<td align="center">3x</td>395<td align="center">0.652</td>396<td align="center">0.145</td>397<td align="center">6.3</td>398<td align="center">42.6</td>399<td align="center">36.7</td>400<td align="center">138363239</td>401<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_101_C4_3x/138363239/model_final_a2914c.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_101_C4_3x/138363239/metrics.json">metrics</a></td>402</tr>403<!-- ROW: mask_rcnn_R_101_DC5_3x -->404 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_R_101_DC5_3x.yaml">R101-DC5</a></td>405<td align="center">3x</td>406<td align="center">0.545</td>407<td align="center">0.092</td>408<td align="center">7.6</td>409<td align="center">41.9</td>410<td align="center">37.3</td>411<td align="center">138363294</td>412<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_101_DC5_3x/138363294/model_final_0464b7.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_101_DC5_3x/138363294/metrics.json">metrics</a></td>413</tr>414<!-- ROW: mask_rcnn_R_101_FPN_3x -->415 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_R_101_FPN_3x.yaml">R101-FPN</a></td>416<td align="center">3x</td>417<td align="center">0.340</td>418<td align="center">0.056</td>419<td align="center">4.6</td>420<td align="center">42.9</td>421<td align="center">38.6</td>422<td align="center">138205316</td>423<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_101_FPN_3x/138205316/model_final_a3ec72.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_101_FPN_3x/138205316/metrics.json">metrics</a></td>424</tr>425<!-- ROW: mask_rcnn_X_101_32x8d_FPN_3x -->426 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_X_101_32x8d_FPN_3x.yaml">X101-FPN</a></td>427<td align="center">3x</td>428<td align="center">0.690</td>429<td align="center">0.103</td>430<td align="center">7.2</td>431<td align="center">44.3</td>432<td align="center">39.5</td>433<td align="center">139653917</td>434<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_X_101_32x8d_FPN_3x/139653917/model_final_2d9806.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_X_101_32x8d_FPN_3x/139653917/metrics.json">metrics</a></td>435</tr>436</tbody></table>437 438 439 440#### New baselines using Large-Scale Jitter and Longer Training Schedule441 442The following baselines of COCO Instance Segmentation with Mask R-CNN are generated443using a longer training schedule and large-scale jitter as described in Google's444[Simple Copy-Paste Data Augmentation](https://arxiv.org/pdf/2012.07177.pdf) paper. These445models are trained from scratch using random initialization. These baselines exceed the446previous Mask R-CNN baselines.447 448In the following table, one epoch consists of training on 118000 COCO images.449 450<table><tbody>451<!-- START TABLE -->452<!-- TABLE HEADER -->453<th valign="bottom">Name</th>454<th valign="bottom">epochs</th>455<th valign="bottom">train<br/>time<br/>(s/im)</th>456<th valign="bottom">inference<br/>time<br/>(s/im)</th>457<th valign="bottom">box<br/>AP</th>458<th valign="bottom">mask<br/>AP</th>459<th valign="bottom">model id</th>460<th valign="bottom">download</th>461<!-- TABLE BODY -->462<!-- ROW: mask_rcnn_R_50_FPN_100ep_LSJ -->463 <tr><td align="left"><a href="configs/new_baselines/mask_rcnn_R_50_FPN_100ep_LSJ.py">R50-FPN</a></td>464<td align="center">100</td>465<td align="center">0.376</td>466<td align="center">0.069</td>467<td align="center">44.6</td>468<td align="center">40.3</td>469<td align="center">42047764</td>470<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_R_50_FPN_100ep_LSJ/42047764/model_final_bb69de.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_R_50_FPN_100ep_LSJ/42047764/metrics.json">metrics</a></td>471</tr>472<!-- ROW: mask_rcnn_R_50_FPN_200ep_LSJ -->473 <tr><td align="left"><a href="configs/new_baselines/mask_rcnn_R_50_FPN_200ep_LSJ.py">R50-FPN</a></td>474<td align="center">200</td>475<td align="center">0.376</td>476<td align="center">0.069</td>477<td align="center">46.3</td>478<td align="center">41.7</td>479<td align="center">42047638</td>480<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_R_50_FPN_200ep_LSJ/42047638/model_final_89a8d3.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_R_50_FPN_200ep_LSJ/42047638/metrics.json">metrics</a></td>481</tr>482<!-- ROW: mask_rcnn_R_50_FPN_400ep_LSJ -->483 <tr><td align="left"><a href="configs/new_baselines/mask_rcnn_R_50_FPN_400ep_LSJ.py">R50-FPN</a></td>484<td align="center">400</td>485<td align="center">0.376</td>486<td align="center">0.069</td>487<td align="center">47.4</td>488<td align="center">42.5</td>489<td align="center">42019571</td>490<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_R_50_FPN_400ep_LSJ/42019571/model_final_14d201.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_R_50_FPN_400ep_LSJ/42019571/metrics.json">metrics</a></td>491</tr>492<!-- ROW: mask_rcnn_R_101_FPN_100ep_LSJ -->493 <tr><td align="left"><a href="configs/new_baselines/mask_rcnn_R_101_FPN_100ep_LSJ.py">R101-FPN</a></td>494<td align="center">100</td>495<td align="center">0.518</td>496<td align="center">0.073</td>497<td align="center">46.4</td>498<td align="center">41.6</td>499<td align="center">42025812</td>500<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_R_101_FPN_100ep_LSJ/42025812/model_final_4f7b58.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_R_101_FPN_100ep_LSJ/42025812/metrics.json">metrics</a></td>501</tr>502<!-- ROW: mask_rcnn_R_101_FPN_200ep_LSJ -->503 <tr><td align="left"><a href="configs/new_baselines/mask_rcnn_R_101_FPN_200ep_LSJ.py">R101-FPN</a></td>504<td align="center">200</td>505<td align="center">0.518</td>506<td align="center">0.073</td>507<td align="center">48.0</td>508<td align="center">43.1</td>509<td align="center">42131867</td>510<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_R_101_FPN_200ep_LSJ/42131867/model_final_0bb7ae.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_R_101_FPN_200ep_LSJ/42131867/metrics.json">metrics</a></td>511</tr>512<!-- ROW: mask_rcnn_R_101_FPN_400ep_LSJ -->513 <tr><td align="left"><a href="configs/new_baselines/mask_rcnn_R_101_FPN_400ep_LSJ.py">R101-FPN</a></td>514<td align="center">400</td>515<td align="center">0.518</td>516<td align="center">0.073</td>517<td align="center">48.9</td>518<td align="center">43.7</td>519<td align="center">42073830</td>520<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_R_101_FPN_400ep_LSJ/42073830/model_final_f96b26.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_R_101_FPN_400ep_LSJ/42073830/metrics.json">metrics</a></td>521</tr>522<!-- ROW: mask_rcnn_regnetx_4gf_dds_FPN_100ep_LSJ -->523 <tr><td align="left"><a href="configs/new_baselines/mask_rcnn_regnetx_4gf_dds_FPN_100ep_LSJ.py">regnetx_4gf_dds_FPN</a></td>524<td align="center">100</td>525<td align="center">0.474</td>526<td align="center">0.071</td>527<td align="center">46.0</td>528<td align="center">41.3</td>529<td align="center">42047771</td>530<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_regnetx_4gf_dds_FPN_100ep_LSJ/42047771/model_final_b7fbab.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_regnetx_4gf_dds_FPN_100ep_LSJ/42047771/metrics.json">metrics</a></td>531</tr>532<!-- ROW: mask_rcnn_regnetx_4gf_dds_FPN_200ep_LSJ -->533 <tr><td align="left"><a href="configs/new_baselines/mask_rcnn_regnetx_4gf_dds_FPN_200ep_LSJ.py">regnetx_4gf_dds_FPN</a></td>534<td align="center">200</td>535<td align="center">0.474</td>536<td align="center">0.071</td>537<td align="center">48.1</td>538<td align="center">43.1</td>539<td align="center">42132721</td>540<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_regnetx_4gf_dds_FPN_200ep_LSJ/42132721/model_final_5d87c1.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_regnetx_4gf_dds_FPN_200ep_LSJ/42132721/metrics.json">metrics</a></td>541</tr>542<!-- ROW: mask_rcnn_regnetx_4gf_dds_FPN_400ep_LSJ -->543 <tr><td align="left"><a href="configs/new_baselines/mask_rcnn_regnetx_4gf_dds_FPN_400ep_LSJ.py">regnetx_4gf_dds_FPN</a></td>544<td align="center">400</td>545<td align="center">0.474</td>546<td align="center">0.071</td>547<td align="center">48.6</td>548<td align="center">43.5</td>549<td align="center">42025447</td>550<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_regnetx_4gf_dds_FPN_400ep_LSJ/42025447/model_final_f1362d.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_regnetx_4gf_dds_FPN_400ep_LSJ/42025447/metrics.json">metrics</a></td>551</tr>552<!-- ROW: mask_rcnn_regnety_4gf_dds_FPN_100ep_LSJ -->553 <tr><td align="left"><a href="configs/new_baselines/mask_rcnn_regnety_4gf_dds_FPN_100ep_LSJ.py">regnety_4gf_dds_FPN</a></td>554<td align="center">100</td>555<td align="center">0.487</td>556<td align="center">0.073</td>557<td align="center">46.1</td>558<td align="center">41.6</td>559<td align="center">42047784</td>560<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_regnety_4gf_dds_FPN_100ep_LSJ/42047784/model_final_6ba57e.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_regnety_4gf_dds_FPN_100ep_LSJ/42047784/metrics.json">metrics</a></td>561</tr>562<!-- ROW: mask_rcnn_regnety_4gf_dds_FPN_200ep_LSJ -->563 <tr><td align="left"><a href="configs/new_baselines/mask_rcnn_regnety_4gf_dds_FPN_200ep_LSJ.py">regnety_4gf_dds_FPN</a></td>564<td align="center">200</td>565<td align="center">0.487</td>566<td align="center">0.072</td>567<td align="center">47.8</td>568<td align="center">43.0</td>569<td align="center">42047642</td>570<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_regnety_4gf_dds_FPN_200ep_LSJ/42047642/model_final_27b9c1.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_regnety_4gf_dds_FPN_200ep_LSJ/42047642/metrics.json">metrics</a></td>571</tr>572<!-- ROW: mask_rcnn_regnety_4gf_dds_FPN_400ep_LSJ -->573 <tr><td align="left"><a href="configs/new_baselines/mask_rcnn_regnety_4gf_dds_FPN_400ep_LSJ.py">regnety_4gf_dds_FPN</a></td>574<td align="center">400</td>575<td align="center">0.487</td>576<td align="center">0.072</td>577<td align="center">48.2</td>578<td align="center">43.3</td>579<td align="center">42045954</td>580<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_regnety_4gf_dds_FPN_400ep_LSJ/42045954/model_final_ef3a80.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/new_baselines/mask_rcnn_regnety_4gf_dds_FPN_400ep_LSJ/42045954/metrics.json">metrics</a></td>581</tr>582</tbody></table>583 584### COCO Person Keypoint Detection Baselines with Keypoint R-CNN585<!--586./gen_html_table.py --config 'COCO-Keypoints/*50*' 'COCO-Keypoints/*101*' --name R50-FPN R50-FPN R101-FPN X101-FPN --fields lr_sched train_speed inference_speed mem box_AP keypoint_AP587-->588 589 590<table><tbody>591<!-- START TABLE -->592<!-- TABLE HEADER -->593<th valign="bottom">Name</th>594<th valign="bottom">lr<br/>sched</th>595<th valign="bottom">train<br/>time<br/>(s/iter)</th>596<th valign="bottom">inference<br/>time<br/>(s/im)</th>597<th valign="bottom">train<br/>mem<br/>(GB)</th>598<th valign="bottom">box<br/>AP</th>599<th valign="bottom">kp.<br/>AP</th>600<th valign="bottom">model id</th>601<th valign="bottom">download</th>602<!-- TABLE BODY -->603<!-- ROW: keypoint_rcnn_R_50_FPN_1x -->604 <tr><td align="left"><a href="configs/COCO-Keypoints/keypoint_rcnn_R_50_FPN_1x.yaml">R50-FPN</a></td>605<td align="center">1x</td>606<td align="center">0.315</td>607<td align="center">0.072</td>608<td align="center">5.0</td>609<td align="center">53.6</td>610<td align="center">64.0</td>611<td align="center">137261548</td>612<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Keypoints/keypoint_rcnn_R_50_FPN_1x/137261548/model_final_04e291.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Keypoints/keypoint_rcnn_R_50_FPN_1x/137261548/metrics.json">metrics</a></td>613</tr>614<!-- ROW: keypoint_rcnn_R_50_FPN_3x -->615 <tr><td align="left"><a href="configs/COCO-Keypoints/keypoint_rcnn_R_50_FPN_3x.yaml">R50-FPN</a></td>616<td align="center">3x</td>617<td align="center">0.316</td>618<td align="center">0.066</td>619<td align="center">5.0</td>620<td align="center">55.4</td>621<td align="center">65.5</td>622<td align="center">137849621</td>623<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Keypoints/keypoint_rcnn_R_50_FPN_3x/137849621/model_final_a6e10b.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Keypoints/keypoint_rcnn_R_50_FPN_3x/137849621/metrics.json">metrics</a></td>624</tr>625<!-- ROW: keypoint_rcnn_R_101_FPN_3x -->626 <tr><td align="left"><a href="configs/COCO-Keypoints/keypoint_rcnn_R_101_FPN_3x.yaml">R101-FPN</a></td>627<td align="center">3x</td>628<td align="center">0.390</td>629<td align="center">0.076</td>630<td align="center">6.1</td>631<td align="center">56.4</td>632<td align="center">66.1</td>633<td align="center">138363331</td>634<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Keypoints/keypoint_rcnn_R_101_FPN_3x/138363331/model_final_997cc7.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Keypoints/keypoint_rcnn_R_101_FPN_3x/138363331/metrics.json">metrics</a></td>635</tr>636<!-- ROW: keypoint_rcnn_X_101_32x8d_FPN_3x -->637 <tr><td align="left"><a href="configs/COCO-Keypoints/keypoint_rcnn_X_101_32x8d_FPN_3x.yaml">X101-FPN</a></td>638<td align="center">3x</td>639<td align="center">0.738</td>640<td align="center">0.121</td>641<td align="center">8.7</td>642<td align="center">57.3</td>643<td align="center">66.0</td>644<td align="center">139686956</td>645<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Keypoints/keypoint_rcnn_X_101_32x8d_FPN_3x/139686956/model_final_5ad38f.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-Keypoints/keypoint_rcnn_X_101_32x8d_FPN_3x/139686956/metrics.json">metrics</a></td>646</tr>647</tbody></table>648 649### COCO Panoptic Segmentation Baselines with Panoptic FPN650<!--651./gen_html_table.py --config 'COCO-PanopticSegmentation/*50*' 'COCO-PanopticSegmentation/*101*' --name R50-FPN R50-FPN R101-FPN --fields lr_sched train_speed inference_speed mem box_AP mask_AP PQ652-->653 654 655<table><tbody>656<!-- START TABLE -->657<!-- TABLE HEADER -->658<th valign="bottom">Name</th>659<th valign="bottom">lr<br/>sched</th>660<th valign="bottom">train<br/>time<br/>(s/iter)</th>661<th valign="bottom">inference<br/>time<br/>(s/im)</th>662<th valign="bottom">train<br/>mem<br/>(GB)</th>663<th valign="bottom">box<br/>AP</th>664<th valign="bottom">mask<br/>AP</th>665<th valign="bottom">PQ</th>666<th valign="bottom">model id</th>667<th valign="bottom">download</th>668<!-- TABLE BODY -->669<!-- ROW: panoptic_fpn_R_50_1x -->670 <tr><td align="left"><a href="configs/COCO-PanopticSegmentation/panoptic_fpn_R_50_1x.yaml">R50-FPN</a></td>671<td align="center">1x</td>672<td align="center">0.304</td>673<td align="center">0.053</td>674<td align="center">4.8</td>675<td align="center">37.6</td>676<td align="center">34.7</td>677<td align="center">39.4</td>678<td align="center">139514544</td>679<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-PanopticSegmentation/panoptic_fpn_R_50_1x/139514544/model_final_dbfeb4.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-PanopticSegmentation/panoptic_fpn_R_50_1x/139514544/metrics.json">metrics</a></td>680</tr>681<!-- ROW: panoptic_fpn_R_50_3x -->682 <tr><td align="left"><a href="configs/COCO-PanopticSegmentation/panoptic_fpn_R_50_3x.yaml">R50-FPN</a></td>683<td align="center">3x</td>684<td align="center">0.302</td>685<td align="center">0.053</td>686<td align="center">4.8</td>687<td align="center">40.0</td>688<td align="center">36.5</td>689<td align="center">41.5</td>690<td align="center">139514569</td>691<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-PanopticSegmentation/panoptic_fpn_R_50_3x/139514569/model_final_c10459.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-PanopticSegmentation/panoptic_fpn_R_50_3x/139514569/metrics.json">metrics</a></td>692</tr>693<!-- ROW: panoptic_fpn_R_101_3x -->694 <tr><td align="left"><a href="configs/COCO-PanopticSegmentation/panoptic_fpn_R_101_3x.yaml">R101-FPN</a></td>695<td align="center">3x</td>696<td align="center">0.392</td>697<td align="center">0.066</td>698<td align="center">6.0</td>699<td align="center">42.4</td>700<td align="center">38.5</td>701<td align="center">43.0</td>702<td align="center">139514519</td>703<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-PanopticSegmentation/panoptic_fpn_R_101_3x/139514519/model_final_cafdb1.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-PanopticSegmentation/panoptic_fpn_R_101_3x/139514519/metrics.json">metrics</a></td>704</tr>705</tbody></table>706 707 708### LVIS Instance Segmentation Baselines with Mask R-CNN709 710Mask R-CNN baselines on the [LVIS dataset](https://lvisdataset.org), v0.5.711These baselines are described in Table 3(c) of the [LVIS paper](https://arxiv.org/abs/1908.03195).712 713NOTE: the 1x schedule here has the same amount of __iterations__ as the COCO 1x baselines.714They are roughly 24 epochs of LVISv0.5 data.715The final results of these configs have large variance across different runs.716 717<!--718./gen_html_table.py --config 'LVISv0.5-InstanceSegmentation/mask*50*' 'LVISv0.5-InstanceSegmentation/mask*101*' --name R50-FPN R101-FPN X101-FPN --fields lr_sched train_speed inference_speed mem box_AP mask_AP719-->720 721 722<table><tbody>723<!-- START TABLE -->724<!-- TABLE HEADER -->725<th valign="bottom">Name</th>726<th valign="bottom">lr<br/>sched</th>727<th valign="bottom">train<br/>time<br/>(s/iter)</th>728<th valign="bottom">inference<br/>time<br/>(s/im)</th>729<th valign="bottom">train<br/>mem<br/>(GB)</th>730<th valign="bottom">box<br/>AP</th>731<th valign="bottom">mask<br/>AP</th>732<th valign="bottom">model id</th>733<th valign="bottom">download</th>734<!-- TABLE BODY -->735<!-- ROW: mask_rcnn_R_50_FPN_1x -->736 <tr><td align="left"><a href="configs/LVISv0.5-InstanceSegmentation/mask_rcnn_R_50_FPN_1x.yaml">R50-FPN</a></td>737<td align="center">1x</td>738<td align="center">0.292</td>739<td align="center">0.107</td>740<td align="center">7.1</td>741<td align="center">23.6</td>742<td align="center">24.4</td>743<td align="center">144219072</td>744<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/LVISv0.5-InstanceSegmentation/mask_rcnn_R_50_FPN_1x/144219072/model_final_571f7c.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/LVISv0.5-InstanceSegmentation/mask_rcnn_R_50_FPN_1x/144219072/metrics.json">metrics</a></td>745</tr>746<!-- ROW: mask_rcnn_R_101_FPN_1x -->747 <tr><td align="left"><a href="configs/LVISv0.5-InstanceSegmentation/mask_rcnn_R_101_FPN_1x.yaml">R101-FPN</a></td>748<td align="center">1x</td>749<td align="center">0.371</td>750<td align="center">0.114</td>751<td align="center">7.8</td>752<td align="center">25.6</td>753<td align="center">25.9</td>754<td align="center">144219035</td>755<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/LVISv0.5-InstanceSegmentation/mask_rcnn_R_101_FPN_1x/144219035/model_final_824ab5.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/LVISv0.5-InstanceSegmentation/mask_rcnn_R_101_FPN_1x/144219035/metrics.json">metrics</a></td>756</tr>757<!-- ROW: mask_rcnn_X_101_32x8d_FPN_1x -->758 <tr><td align="left"><a href="configs/LVISv0.5-InstanceSegmentation/mask_rcnn_X_101_32x8d_FPN_1x.yaml">X101-FPN</a></td>759<td align="center">1x</td>760<td align="center">0.712</td>761<td align="center">0.151</td>762<td align="center">10.2</td>763<td align="center">26.7</td>764<td align="center">27.1</td>765<td align="center">144219108</td>766<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/LVISv0.5-InstanceSegmentation/mask_rcnn_X_101_32x8d_FPN_1x/144219108/model_final_5e3439.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/LVISv0.5-InstanceSegmentation/mask_rcnn_X_101_32x8d_FPN_1x/144219108/metrics.json">metrics</a></td>767</tr>768</tbody></table>769 770 771 772### Cityscapes & Pascal VOC Baselines773 774Simple baselines for775* Mask R-CNN on Cityscapes instance segmentation (initialized from COCO pre-training, then trained on Cityscapes fine annotations only)776* Faster R-CNN on PASCAL VOC object detection (trained on VOC 2007 train+val + VOC 2012 train+val, tested on VOC 2007 using 11-point interpolated AP)777 778<!--779./gen_html_table.py --config 'Cityscapes/*' 'PascalVOC-Detection/*' --name "R50-FPN, Cityscapes" "R50-C4, VOC" --fields train_speed inference_speed mem box_AP box_AP50 mask_AP780-->781 782 783<table><tbody>784<!-- START TABLE -->785<!-- TABLE HEADER -->786<th valign="bottom">Name</th>787<th valign="bottom">train<br/>time<br/>(s/iter)</th>788<th valign="bottom">inference<br/>time<br/>(s/im)</th>789<th valign="bottom">train<br/>mem<br/>(GB)</th>790<th valign="bottom">box<br/>AP</th>791<th valign="bottom">box<br/>AP50</th>792<th valign="bottom">mask<br/>AP</th>793<th valign="bottom">model id</th>794<th valign="bottom">download</th>795<!-- TABLE BODY -->796<!-- ROW: mask_rcnn_R_50_FPN -->797 <tr><td align="left"><a href="configs/Cityscapes/mask_rcnn_R_50_FPN.yaml">R50-FPN, Cityscapes</a></td>798<td align="center">0.240</td>799<td align="center">0.078</td>800<td align="center">4.4</td>801<td align="center"></td>802<td align="center"></td>803<td align="center">36.5</td>804<td align="center">142423278</td>805<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/Cityscapes/mask_rcnn_R_50_FPN/142423278/model_final_af9cf5.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/Cityscapes/mask_rcnn_R_50_FPN/142423278/metrics.json">metrics</a></td>806</tr>807<!-- ROW: faster_rcnn_R_50_C4 -->808 <tr><td align="left"><a href="configs/PascalVOC-Detection/faster_rcnn_R_50_C4.yaml">R50-C4, VOC</a></td>809<td align="center">0.537</td>810<td align="center">0.081</td>811<td align="center">4.8</td>812<td align="center">51.9</td>813<td align="center">80.3</td>814<td align="center"></td>815<td align="center">142202221</td>816<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/PascalVOC-Detection/faster_rcnn_R_50_C4/142202221/model_final_b1acc2.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/PascalVOC-Detection/faster_rcnn_R_50_C4/142202221/metrics.json">metrics</a></td>817</tr>818</tbody></table>819 820 821 822### Other Settings823 824Ablations for Deformable Conv and Cascade R-CNN:825 826<!--827./gen_html_table.py --config 'COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_1x.yaml' 'Misc/*R_50_FPN_1x_dconv*' 'Misc/cascade*1x.yaml' 'COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml' 'Misc/*R_50_FPN_3x_dconv*' 'Misc/cascade*3x.yaml' --name "Baseline R50-FPN" "Deformable Conv" "Cascade R-CNN" "Baseline R50-FPN" "Deformable Conv" "Cascade R-CNN" --fields lr_sched train_speed inference_speed mem box_AP mask_AP828-->829 830 831<table><tbody>832<!-- START TABLE -->833<!-- TABLE HEADER -->834<th valign="bottom">Name</th>835<th valign="bottom">lr<br/>sched</th>836<th valign="bottom">train<br/>time<br/>(s/iter)</th>837<th valign="bottom">inference<br/>time<br/>(s/im)</th>838<th valign="bottom">train<br/>mem<br/>(GB)</th>839<th valign="bottom">box<br/>AP</th>840<th valign="bottom">mask<br/>AP</th>841<th valign="bottom">model id</th>842<th valign="bottom">download</th>843<!-- TABLE BODY -->844<!-- ROW: mask_rcnn_R_50_FPN_1x -->845 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_1x.yaml">Baseline R50-FPN</a></td>846<td align="center">1x</td>847<td align="center">0.261</td>848<td align="center">0.043</td>849<td align="center">3.4</td>850<td align="center">38.6</td>851<td align="center">35.2</td>852<td align="center">137260431</td>853<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_1x/137260431/model_final_a54504.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_1x/137260431/metrics.json">metrics</a></td>854</tr>855<!-- ROW: mask_rcnn_R_50_FPN_1x_dconv_c3-c5 -->856 <tr><td align="left"><a href="configs/Misc/mask_rcnn_R_50_FPN_1x_dconv_c3-c5.yaml">Deformable Conv</a></td>857<td align="center">1x</td>858<td align="center">0.342</td>859<td align="center">0.048</td>860<td align="center">3.5</td>861<td align="center">41.5</td>862<td align="center">37.5</td>863<td align="center">138602867</td>864<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/Misc/mask_rcnn_R_50_FPN_1x_dconv_c3-c5/138602867/model_final_65c703.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/Misc/mask_rcnn_R_50_FPN_1x_dconv_c3-c5/138602867/metrics.json">metrics</a></td>865</tr>866<!-- ROW: cascade_mask_rcnn_R_50_FPN_1x -->867 <tr><td align="left"><a href="configs/Misc/cascade_mask_rcnn_R_50_FPN_1x.yaml">Cascade R-CNN</a></td>868<td align="center">1x</td>869<td align="center">0.317</td>870<td align="center">0.052</td>871<td align="center">4.0</td>872<td align="center">42.1</td>873<td align="center">36.4</td>874<td align="center">138602847</td>875<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/Misc/cascade_mask_rcnn_R_50_FPN_1x/138602847/model_final_e9d89b.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/Misc/cascade_mask_rcnn_R_50_FPN_1x/138602847/metrics.json">metrics</a></td>876</tr>877<!-- ROW: mask_rcnn_R_50_FPN_3x -->878 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml">Baseline R50-FPN</a></td>879<td align="center">3x</td>880<td align="center">0.261</td>881<td align="center">0.043</td>882<td align="center">3.4</td>883<td align="center">41.0</td>884<td align="center">37.2</td>885<td align="center">137849600</td>886<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x/137849600/model_final_f10217.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x/137849600/metrics.json">metrics</a></td>887</tr>888<!-- ROW: mask_rcnn_R_50_FPN_3x_dconv_c3-c5 -->889 <tr><td align="left"><a href="configs/Misc/mask_rcnn_R_50_FPN_3x_dconv_c3-c5.yaml">Deformable Conv</a></td>890<td align="center">3x</td>891<td align="center">0.349</td>892<td align="center">0.047</td>893<td align="center">3.5</td>894<td align="center">42.7</td>895<td align="center">38.5</td>896<td align="center">144998336</td>897<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/Misc/mask_rcnn_R_50_FPN_3x_dconv_c3-c5/144998336/model_final_821d0b.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/Misc/mask_rcnn_R_50_FPN_3x_dconv_c3-c5/144998336/metrics.json">metrics</a></td>898</tr>899<!-- ROW: cascade_mask_rcnn_R_50_FPN_3x -->900 <tr><td align="left"><a href="configs/Misc/cascade_mask_rcnn_R_50_FPN_3x.yaml">Cascade R-CNN</a></td>901<td align="center">3x</td>902<td align="center">0.328</td>903<td align="center">0.053</td>904<td align="center">4.0</td>905<td align="center">44.3</td>906<td align="center">38.5</td>907<td align="center">144998488</td>908<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/Misc/cascade_mask_rcnn_R_50_FPN_3x/144998488/model_final_480dd8.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/Misc/cascade_mask_rcnn_R_50_FPN_3x/144998488/metrics.json">metrics</a></td>909</tr>910</tbody></table>911 912 913Ablations for normalization methods, and a few models trained from scratch following [Rethinking ImageNet Pre-training](https://arxiv.org/abs/1811.08883).914(Note: The baseline uses `2fc` head while the others use [`4conv1fc` head](https://arxiv.org/abs/1803.08494))915<!--916./gen_html_table.py --config 'COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml' 'Misc/mask*50_FPN_3x_gn.yaml' 'Misc/mask*50_FPN_3x_syncbn.yaml' 'Misc/scratch*' --name "Baseline R50-FPN" "GN" "SyncBN" "GN (from scratch)" "GN (from scratch)" "SyncBN (from scratch)" --fields lr_sched train_speed inference_speed mem box_AP mask_AP917 -->918 919 920<table><tbody>921<!-- START TABLE -->922<!-- TABLE HEADER -->923<th valign="bottom">Name</th>924<th valign="bottom">lr<br/>sched</th>925<th valign="bottom">train<br/>time<br/>(s/iter)</th>926<th valign="bottom">inference<br/>time<br/>(s/im)</th>927<th valign="bottom">train<br/>mem<br/>(GB)</th>928<th valign="bottom">box<br/>AP</th>929<th valign="bottom">mask<br/>AP</th>930<th valign="bottom">model id</th>931<th valign="bottom">download</th>932<!-- TABLE BODY -->933<!-- ROW: mask_rcnn_R_50_FPN_3x -->934 <tr><td align="left"><a href="configs/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml">Baseline R50-FPN</a></td>935<td align="center">3x</td>936<td align="center">0.261</td>937<td align="center">0.043</td>938<td align="center">3.4</td>939<td align="center">41.0</td>940<td align="center">37.2</td>941<td align="center">137849600</td>942<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x/137849600/model_final_f10217.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x/137849600/metrics.json">metrics</a></td>943</tr>944<!-- ROW: mask_rcnn_R_50_FPN_3x_gn -->945 <tr><td align="left"><a href="configs/Misc/mask_rcnn_R_50_FPN_3x_gn.yaml">GN</a></td>946<td align="center">3x</td>947<td align="center">0.309</td>948<td align="center">0.060</td>949<td align="center">5.6</td>950<td align="center">42.6</td>951<td align="center">38.6</td>952<td align="center">138602888</td>953<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/Misc/mask_rcnn_R_50_FPN_3x_gn/138602888/model_final_dc5d9e.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/Misc/mask_rcnn_R_50_FPN_3x_gn/138602888/metrics.json">metrics</a></td>954</tr>955<!-- ROW: mask_rcnn_R_50_FPN_3x_syncbn -->956 <tr><td align="left"><a href="configs/Misc/mask_rcnn_R_50_FPN_3x_syncbn.yaml">SyncBN</a></td>957<td align="center">3x</td>958<td align="center">0.345</td>959<td align="center">0.053</td>960<td align="center">5.5</td>961<td align="center">41.9</td>962<td align="center">37.8</td>963<td align="center">169527823</td>964<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/Misc/mask_rcnn_R_50_FPN_3x_syncbn/169527823/model_final_3b3c51.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/Misc/mask_rcnn_R_50_FPN_3x_syncbn/169527823/metrics.json">metrics</a></td>965</tr>966<!-- ROW: scratch_mask_rcnn_R_50_FPN_3x_gn -->967 <tr><td align="left"><a href="configs/Misc/scratch_mask_rcnn_R_50_FPN_3x_gn.yaml">GN (from scratch)</a></td>968<td align="center">3x</td>969<td align="center">0.338</td>970<td align="center">0.061</td>971<td align="center">7.2</td>972<td align="center">39.9</td>973<td align="center">36.6</td>974<td align="center">138602908</td>975<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/Misc/scratch_mask_rcnn_R_50_FPN_3x_gn/138602908/model_final_01ca85.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/Misc/scratch_mask_rcnn_R_50_FPN_3x_gn/138602908/metrics.json">metrics</a></td>976</tr>977<!-- ROW: scratch_mask_rcnn_R_50_FPN_9x_gn -->978 <tr><td align="left"><a href="configs/Misc/scratch_mask_rcnn_R_50_FPN_9x_gn.yaml">GN (from scratch)</a></td>979<td align="center">9x</td>980<td align="center">N/A</td>981<td align="center">0.061</td>982<td align="center">7.2</td>983<td align="center">43.7</td>984<td align="center">39.6</td>985<td align="center">183808979</td>986<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/Misc/scratch_mask_rcnn_R_50_FPN_9x_gn/183808979/model_final_da7b4c.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/Misc/scratch_mask_rcnn_R_50_FPN_9x_gn/183808979/metrics.json">metrics</a></td>987</tr>988<!-- ROW: scratch_mask_rcnn_R_50_FPN_9x_syncbn -->989 <tr><td align="left"><a href="configs/Misc/scratch_mask_rcnn_R_50_FPN_9x_syncbn.yaml">SyncBN (from scratch)</a></td>990<td align="center">9x</td>991<td align="center">N/A</td>992<td align="center">0.055</td>993<td align="center">7.2</td>994<td align="center">43.6</td>995<td align="center">39.3</td>996<td align="center">184226666</td>997<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/Misc/scratch_mask_rcnn_R_50_FPN_9x_syncbn/184226666/model_final_5ce33e.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/Misc/scratch_mask_rcnn_R_50_FPN_9x_syncbn/184226666/metrics.json">metrics</a></td>998</tr>999</tbody></table>1000 1001 1002A few very large models trained for a long time, for demo purposes. They are trained using multiple machines:1003 1004<!--1005./gen_html_table.py --config 'Misc/panoptic_*dconv*' 'Misc/cascade_*152*' --name "Panoptic FPN R101" "Mask R-CNN X152" --fields inference_speed mem box_AP mask_AP PQ1006# manually add TTA results1007-->1008 1009 1010<table><tbody>1011<!-- START TABLE -->1012<!-- TABLE HEADER -->1013<th valign="bottom">Name</th>1014<th valign="bottom">inference<br/>time<br/>(s/im)</th>1015<th valign="bottom">train<br/>mem<br/>(GB)</th>1016<th valign="bottom">box<br/>AP</th>1017<th valign="bottom">mask<br/>AP</th>1018<th valign="bottom">PQ</th>1019<th valign="bottom">model id</th>1020<th valign="bottom">download</th>1021<!-- TABLE BODY -->1022<!-- ROW: panoptic_fpn_R_101_dconv_cascade_gn_3x -->1023 <tr><td align="left"><a href="configs/Misc/panoptic_fpn_R_101_dconv_cascade_gn_3x.yaml">Panoptic FPN R101</a></td>1024<td align="center">0.098</td>1025<td align="center">11.4</td>1026<td align="center">47.4</td>1027<td align="center">41.3</td>1028<td align="center">46.1</td>1029<td align="center">139797668</td>1030<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/Misc/panoptic_fpn_R_101_dconv_cascade_gn_3x/139797668/model_final_be35db.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/Misc/panoptic_fpn_R_101_dconv_cascade_gn_3x/139797668/metrics.json">metrics</a></td>1031</tr>1032<!-- ROW: cascade_mask_rcnn_X_152_32x8d_FPN_IN5k_gn_dconv -->1033 <tr><td align="left"><a href="configs/Misc/cascade_mask_rcnn_X_152_32x8d_FPN_IN5k_gn_dconv.yaml">Mask R-CNN X152</a></td>1034<td align="center">0.234</td>1035<td align="center">15.1</td>1036<td align="center">50.2</td>1037<td align="center">44.0</td>1038<td align="center"></td>1039<td align="center">18131413</td>1040<td align="center"><a href="https://dl.fbaipublicfiles.com/detectron2/Misc/cascade_mask_rcnn_X_152_32x8d_FPN_IN5k_gn_dconv/18131413/model_0039999_e76410.pkl">model</a> | <a href="https://dl.fbaipublicfiles.com/detectron2/Misc/cascade_mask_rcnn_X_152_32x8d_FPN_IN5k_gn_dconv/18131413/metrics.json">metrics</a></td>1041</tr>1042<!-- ROW: TTA cascade_mask_rcnn_X_152_32x8d_FPN_IN5k_gn_dconv -->1043 <tr><td align="left">above + test-time aug.</td>1044<td align="center"></td>1045<td align="center"></td>1046<td align="center">51.9</td>1047<td align="center">45.9</td>1048<td align="center"></td>1049<td align="center"></td>1050<td align="center"></td>1051</tr>1052</tbody></table>1053