Rename RTMPose-t to DOPER-t, add HOPE results (28 objects)

README.md

@@ -3,37 +3,45 @@ license: apache-2.0
 tags:
 - pose-estimation
 - bop
-- rtmpose
+- doper
 - keypoints
 ---
 
-# DOPER BOP -- RTMPose-t Keypoint Models for BOP Core Datasets
+# DOPER BOP -- DOPER-t Keypoint Models for BOP Datasets
 
-Per-object RTMPose-t models trained on BOP BlenderProc PBR data with symmetry-aware 3D keypoints.
-Evaluated on BOP19 test splits using GT bounding boxes, RTMPose inference, and PnP pose estimation.
+Per-object DOPER-t models trained on synthetic PBR data with symmetry-aware 3D keypoints.
+Evaluated on BOP test/val splits using GT bounding boxes, DOPER-t keypoint inference, and PnP pose estimation.
 
 ## Summary
 
-| Dataset | Objects | GT Instances | PnP Solved | ADD-AUC | ADDS-AUC | MSSD-AUC |
-|---------|---------|-------------|------------|---------|----------|----------|
-| **lm** | 15 | 3000 | 3000 | 0.9812 | 0.9917 | 0.9731 |
-| **lmo** | 8 | 1517 | 1515 | 0.9680 | 0.9818 | 0.9592 |
-| **tless** | 30 | 6900 | 6900 | 0.6807 | 0.7018 | 0.6658 |
-| **tudl** | 3 | 600 | 600 | 0.9670 | 0.9822 | 0.9486 |
-| **ycbv** | 21 | 4125 | 4125 | 0.8762 | 0.8975 | 0.8611 |
-| **hb** | 33 | -- | -- | -- | -- | -- |
-| **itodd** | 28 | -- | -- | -- | -- | -- |
-| **icbin** | 2 | 2250 | 2250 | 0.8250 | 0.8522 | 0.7971 |
-| **Overall** | **140** | **18392** | **18390** | **0.8333** | **0.8517** | **0.8198** |
+| Dataset | Objects | Split | GT Instances | PnP Solved | ADD-AUC | ADDS-AUC | MSSD-AUC |
+|---------|---------|-------|-------------|------------|---------|----------|----------|
+| **lm** | 15 | test | 3000 | 3000 | 0.9812 | 0.9917 | 0.9731 |
+| **lmo** | 8 | test | 1517 | 1515 | 0.9680 | 0.9818 | 0.9592 |
+| **tless** | 30 | test_primesense | 6900 | 6900 | 0.6807 | 0.7018 | 0.6658 |
+| **tudl** | 3 | test | 600 | 600 | 0.9670 | 0.9822 | 0.9486 |
+| **ycbv** | 21 | test | 4125 | 4125 | 0.8762 | 0.8975 | 0.8611 |
+| **hb** | 33 | test_primesense | -- | -- | -- | -- | -- |
+| **itodd** | 28 | test | -- | -- | -- | -- | -- |
+| **icbin** | 2 | test | 2250 | 2250 | 0.8250 | 0.8522 | 0.7971 |
+| **hope** | 28 | val | 920 | 920 | 0.8813 | 0.8999 | 0.8651 |
+| **Overall** | **168** | | **19312** | **19310** | **0.8459** | **0.8643** | **0.8317** |
 
 > **hb** and **itodd** test sets do not include public ground truth -- results must be submitted to the [BOP evaluation server](https://bop.felk.cvut.cz/).
+>
+> **hope** is evaluated on the val split (test split has no public GT).
 
 ## Method
 
 1. Generate symmetry-aware 3D keypoints (17 per object) from BOP model meshes + `models_info.json`
-2. Project keypoints onto BlenderProc PBR training images using GT poses
-3. Train RTMPose-t (CSPNeXt-tiny backbone, 256x256 input, 300 epochs, 4x L40S GPUs)
-4. At inference: detect 2D keypoints with RTMPose, solve 6DoF pose via PnP+RANSAC
+2. Project keypoints onto synthetic PBR training images using GT poses
+3. Train DOPER-t (CSPNeXt-tiny backbone, 256x256 input, 300 epochs, 4x L40S GPUs)
+4. At inference: detect 2D keypoints with DOPER-t, solve 6DoF pose via PnP+RANSAC
+
+### Training data
+
+- **BOP core datasets** (lm, lmo, tless, tudl, ycbv, hb, itodd, icbin): BOP BlenderProc PBR (`train_pbr`)
+- **HOPE**: Custom BlenderProc synthetic renders (~40K images per object)
 
 ### Symmetry-aware keypoints
 
@@ -233,11 +241,45 @@ Evaluated on BOP19 test splits using GT bounding boxes, RTMPose inference, and P
 |  2 | -- | 220.6 | 450 | 450 | 0.8396 | 0.8624 | 0.8031 | 0.011 | 0.042 | 30 |
 | **mean** | | | | | **0.8250** | **0.8522** | **0.7971** | | | |
 
+## HOPE (28 objects)
+
+| ID | Sym | Diam (mm) | N GT | Solved | ADD-AUC | ADDS-AUC | MSSD-AUC | ADD<10% | ADDS<10% | Epoch |
+|----|-----|-----------|------|--------|---------|----------|----------|---------|----------|-------|
+|  1 | -- | 107.9 | 45 | 45 | 0.9017 | 0.9172 | 0.8928 | 0.267 | 0.400 | 200 |
+|  2 | -- | 153.9 | 45 | 45 | 0.9211 | 0.9389 | 0.8956 | 0.178 | 0.333 | 300 |
+|  3 | -- | 115.0 | 45 | 45 | 0.8372 | 0.8522 | 0.8239 | 0.356 | 0.422 | 90 |
+|  4 | -- | 89.6 | 20 | 20 | 0.9513 | 0.9688 | 0.9412 | 0.100 | 0.400 | 260 |
+|  5 | -- | 98.0 | 30 | 30 | 0.8267 | 0.8417 | 0.8125 | 0.367 | 0.433 | 220 |
+|  6 | -- | 208.7 | 25 | 25 | 0.9940 | 0.9990 | 0.9920 | 0.320 | 0.720 | 300 |
+|  7 | -- | 89.6 | 40 | 40 | 0.8538 | 0.8900 | 0.8300 | 0.075 | 0.175 | 140 |
+|  8 | -- | 115.6 | 25 | 25 | 0.9290 | 0.9480 | 0.9080 | 0.200 | 0.360 | 30 |
+|  9 | -- | 206.0 | 25 | 25 | 0.8520 | 0.8780 | 0.8300 | 0.080 | 0.360 | 120 |
+| 10 | -- | 89.7 | 35 | 35 | 0.5721 | 0.5957 | 0.5471 | 0.114 | 0.314 | 80 |
+| 11 | -- | 153.9 | 40 | 40 | 0.9338 | 0.9444 | 0.9244 | 0.325 | 0.575 | 80 |
+| 12 | -- | 207.1 | 50 | 50 | 0.9300 | 0.9525 | 0.8950 | 0.220 | 0.420 | 180 |
+| 13 | -- | 153.4 | 30 | 30 | 0.9742 | 0.9883 | 0.9617 | 0.200 | 0.333 | 230 |
+| 14 | -- | 204.5 | 25 | 25 | 0.9890 | 0.9970 | 0.9790 | 0.160 | 0.520 | 100 |
+| 15 | -- | 75.7 | 35 | 35 | 0.8907 | 0.9279 | 0.8679 | 0.000 | 0.029 | 90 |
+| 16 | -- | 161.5 | 50 | 50 | 0.9425 | 0.9515 | 0.9300 | 0.440 | 0.640 | 200 |
+| 17 | -- | 205.7 | 25 | 25 | 0.7170 | 0.7320 | 0.6780 | 0.160 | 0.280 | 230 |
+| 18 | -- | 122.8 | 30 | 30 | 0.9300 | 0.9408 | 0.9192 | 0.500 | 0.700 | 210 |
+| 19 | -- | 89.2 | 20 | 20 | 0.9587 | 0.9750 | 0.9550 | 0.150 | 0.300 | 70 |
+| 20 | -- | 89.9 | 30 | 30 | 0.7617 | 0.7933 | 0.7433 | 0.100 | 0.100 | 100 |
+| 21 | -- | 89.2 | 55 | 55 | 0.8659 | 0.8877 | 0.8509 | 0.182 | 0.327 | 270 |
+| 22 | -- | 152.4 | 20 | 20 | 0.8100 | 0.8375 | 0.7900 | 0.000 | 0.050 | 290 |
+| 23 | -- | 151.3 | 40 | 40 | 0.8850 | 0.9125 | 0.8525 | 0.075 | 0.225 | 180 |
+| 24 | -- | 151.3 | 10 | 10 | 0.9675 | 0.9725 | 0.9525 | 0.400 | 0.400 | 260 |
+| 25 | -- | 252.8 | 35 | 35 | 0.9793 | 0.9886 | 0.9736 | 0.486 | 0.571 | 220 |
+| 26 | -- | 107.1 | 35 | 35 | 0.7779 | 0.7929 | 0.7664 | 0.286 | 0.429 | 290 |
+| 27 | -- | 76.1 | 35 | 35 | 0.7586 | 0.7857 | 0.7514 | 0.000 | 0.143 | 60 |
+| 28 | -- | 82.9 | 20 | 20 | 0.9663 | 0.9875 | 0.9587 | 0.100 | 0.250 | 190 |
+| **mean** | | | | | **0.8813** | **0.8999** | **0.8651** | | | |
+
 ## File Structure
 
 ```
 {dataset}/obj_{NNNNNN}/
-  best_coco_AP_epoch_NNN.pth   # RTMPose-t checkpoint
+  best_coco_AP_epoch_NNN.pth   # DOPER-t checkpoint
   keypoints_3d.json             # 17 symmetry-aware 3D keypoints (mm)
   bop_summary.json              # Evaluation metrics
   vis_grid.jpg                  # Qualitative results