Team Ai
Modelpublic

cmes-deepvision/ACR-cuboid-v1.1.1

sourceHugging Faceapache-2.0updated 3mo agoView on Hugging Face
0likes7downloads
Model Card

ACR-cuboid-v1.1.1

Single-class (carton box) 3D cuboid pose estimation model for ACR robotic picking.

Changes from v1.1.0

  • —Bugfix: Fixed 3 geometric augmentation bugs in rgbd_transforms.py:
  • —RandomRotation90RGBD: k=1 / k=3 center_2d and box label formulas were swapped
  • —RandomRotateRGBD: box width/height not updated after rotation
  • —MultiCropRGBD: stale labels returned when no objects survive crop
  • —Retrained 30 epochs with full geometric augmentation pipeline (v1.1.0 used paper-only augmentation)

Eval (checkpoint-2100, epoch 30)

MetricValTest
3D IoU AP@0.5099.97%100.0%
3D IoU AP@0.7594.85%93.46%
COCO 2D AP99.76%99.45%
F1 (score≥0.5)99.65%99.83%
Translation (median)10.28mm10.33mm
Rotation (median)1.41°1.24°
Size LWH (median)12.53mm12.34mm

Architecture

  • —OpenVocabCuboidMapAnythingDINOFor3DObjectDetection
  • —Backbone: DINOv2-S (384-dim, 12 layers, registers)
  • —Decoder: 9-layer deformable DETR
  • —Parameters: 122.4M
  • —Input: 1280×1280

Compatibility

Drop-in replacement for v1.1.0 — identical config, architecture, and weight keys.