[Bug] Can not reproduce the hand-keypoint accuracy
Prerequisite
- I have searched Issues and Discussions but cannot get the expected help.
- The bug has not been fixed in the latest version(https://github.com/open-mmlab/mmpose).
Environment
Package Version Editable project location
addict 2.4.0 albucore 0.0.17 albumentations 1.4.18 aliyun-python-sdk-core 2.16.0 aliyun-python-sdk-kms 2.16.5 annotated-types 0.7.0 certifi 2025.10.5 cffi 1.17.1 charset-normalizer 3.4.4 chumpy 0.70 click 8.1.8 colorama 0.4.6 contourpy 1.1.1 crcmod 1.7 cryptography 46.0.3 cycler 0.12.1 Cython 3.2.0 einops 0.8.1 eval_type_backport 0.2.2 filelock 3.14.0 fonttools 4.57.0 idna 3.11 imageio 2.35.1 importlib_metadata 8.5.0 importlib_resources 6.4.5 jmespath 0.10.0 json-tricks 3.17.3 kiwisolver 1.4.7 lazy_loader 0.4 Markdown 3.7 markdown-it-py 3.0.0 matplotlib 3.7.5 mdurl 0.1.2 mmcv 2.1.0 mmdet 3.3.0 mmengine 0.10.7 mmpose 1.3.2 /home/weichaoyu/mmpose_test model-index 0.1.11 munkres 1.1.4 networkx 3.1 numpy 1.24.4 opencv-python 4.12.0.88 opencv-python-headless 4.12.0.88 opendatalab 0.0.10 openmim 0.3.9 openxlab 0.1.3 ordered-set 4.1.0 oss2 2.17.0 packaging 24.2 pandas 2.0.3 pillow 10.4.0 pip 22.3.1 platformdirs 4.3.6 pycocotools 2.0.7 pycparser 2.23 pycryptodome 3.23.0 pydantic 2.10.6 pydantic_core 2.27.2 Pygments 2.19.2 pyparsing 3.1.4 python-dateutil 2.9.0.post0 pytz 2023.4 PyWavelets 1.4.1 PyYAML 6.0.3 requests 2.28.2 rich 13.4.2 scikit-image 0.21.0 scipy 1.10.1 setuptools 60.2.0 shapely 2.0.7 six 1.17.0 tabulate 0.9.0 termcolor 2.4.0 terminaltables 3.1.10 tifffile 2023.7.10 tomli 2.3.0 torch 1.12.1+cu113 torchaudio 0.12.1+cu113 torchvision 0.13.1+cu113 tqdm 4.65.2 typing_extensions 4.13.2 tzdata 2025.2 urllib3 1.26.20 wheel 0.38.4 xtcocotools 1.14.3 yapf 0.43.0 zipp 3.20.2
Reproduces the problem - code sample
I can not reproduce the hand-keypoint accuracy.
I trained it by
tools/dist_train.sh configs/hand_2d_keypoint/rtmpose/coco_wholebody_hand/rtmpose-m_8xb32-210e_coco-wholebody-hand-256x256.py 1
the result is
2025/11/11 07:15:49 - mmengine - INFO - Exp name: rtmpose-m_8xb32-210e_coco-wholebody-hand-256x256_20251111_041351
2025/11/11 07:16:03 - mmengine - INFO - Epoch(train) [210][ 50/1232] base_lr: 2.007841e-04 lr: 2.007841e-04 eta: 0:05:49 time: 0.270155 data_time: 0.074107 memory: 3062 loss: 0.165204 loss_kpt: 0.165204 acc_pose: 0.315476
2025/11/11 07:16:15 - mmengine - INFO - Epoch(train) [210][ 100/1232] base_lr: 2.007192e-04 lr: 2.007192e-04 eta: 0:05:34 time: 0.251002 data_time: 0.054464 memory: 3062 loss: 0.171634 loss_kpt: 0.171634 acc_pose: 0.289449
2025/11/11 07:16:28 - mmengine - INFO - Epoch(train) [210][ 150/1232] base_lr: 2.006571e-04 lr: 2.006571e-04 eta: 0:05:19 time: 0.244595 data_time: 0.053222 memory: 3062 loss: 0.166850 loss_kpt: 0.166850 acc_pose: 0.282293
2025/11/11 07:16:40 - mmengine - INFO - Epoch(train) [210][ 200/1232] base_lr: 2.005979e-04 lr: 2.005979e-04 eta: 0:05:04 time: 0.247182 data_time: 0.054206 memory: 3062 loss: 0.173348 loss_kpt: 0.173348 acc_pose: 0.317252
2025/11/11 07:16:53 - mmengine - INFO - Epoch(train) [210][ 250/1232] base_lr: 2.005414e-04 lr: 2.005414e-04 eta: 0:04:49 time: 0.249978 data_time: 0.057801 memory: 3062 loss: 0.168208 loss_kpt: 0.168208 acc_pose: 0.318820
2025/11/11 07:17:05 - mmengine - INFO - Epoch(train) [210][ 300/1232] base_lr: 2.004877e-04 lr: 2.004877e-04 eta: 0:04:35 time: 0.237039 data_time: 0.051933 memory: 3062 loss: 0.163171 loss_kpt: 0.163171 acc_pose: 0.312596
2025/11/11 07:17:17 - mmengine - INFO - Epoch(train) [210][ 350/1232] base_lr: 2.004368e-04 lr: 2.004368e-04 eta: 0:04:20 time: 0.249114 data_time: 0.057370 memory: 3062 loss: 0.170330 loss_kpt: 0.170330 acc_pose: 0.358967
2025/11/11 07:17:29 - mmengine - INFO - Epoch(train) [210][ 400/1232] base_lr: 2.003888e-04 lr: 2.003888e-04 eta: 0:04:05 time: 0.239909 data_time: 0.052413 memory: 3062 loss: 0.174126 loss_kpt: 0.174126 acc_pose: 0.231471
2025/11/11 07:17:41 - mmengine - INFO - Epoch(train) [210][ 450/1232] base_lr: 2.003435e-04 lr: 2.003435e-04 eta: 0:03:50 time: 0.244229 data_time: 0.056419 memory: 3062 loss: 0.161479 loss_kpt: 0.161479 acc_pose: 0.303859
2025/11/11 07:17:54 - mmengine - INFO - Epoch(train) [210][ 500/1232] base_lr: 2.003010e-04 lr: 2.003010e-04 eta: 0:03:35 time: 0.246709 data_time: 0.055326 memory: 3062 loss: 0.169996 loss_kpt: 0.169996 acc_pose: 0.289075
2025/11/11 07:17:59 - mmengine - INFO - Exp name: rtmpose-m_8xb32-210e_coco-wholebody-hand-256x256_20251111_041351
2025/11/11 07:18:10 - mmengine - INFO - Epoch(train) [210][ 550/1232] base_lr: 2.002614e-04 lr: 2.002614e-04 eta: 0:03:21 time: 0.322410 data_time: 0.105040 memory: 3062 loss: 0.173382 loss_kpt: 0.173382 acc_pose: 0.327765
2025/11/11 07:18:23 - mmengine - INFO - Epoch(train) [210][ 600/1232] base_lr: 2.002245e-04 lr: 2.002245e-04 eta: 0:03:06 time: 0.259074 data_time: 0.058977 memory: 3062 loss: 0.168148 loss_kpt: 0.168148 acc_pose: 0.263345
2025/11/11 07:18:35 - mmengine - INFO - Epoch(train) [210][ 650/1232] base_lr: 2.001904e-04 lr: 2.001904e-04 eta: 0:02:51 time: 0.244092 data_time: 0.054522 memory: 3062 loss: 0.173146 loss_kpt: 0.173146 acc_pose: 0.306925
2025/11/11 07:18:48 - mmengine - INFO - Epoch(train) [210][ 700/1232] base_lr: 2.001592e-04 lr: 2.001592e-04 eta: 0:02:36 time: 0.253222 data_time: 0.063943 memory: 3062 loss: 0.176185 loss_kpt: 0.176185 acc_pose: 0.335877
2025/11/11 07:19:03 - mmengine - INFO - Epoch(train) [210][ 750/1232] base_lr: 2.001307e-04 lr: 2.001307e-04 eta: 0:02:22 time: 0.300577 data_time: 0.088285 memory: 3062 loss: 0.170724 loss_kpt: 0.170724 acc_pose: 0.367515
2025/11/11 07:19:17 - mmengine - INFO - Epoch(train) [210][ 800/1232] base_lr: 2.001050e-04 lr: 2.001050e-04 eta: 0:02:07 time: 0.277363 data_time: 0.071170 memory: 3062 loss: 0.168807 loss_kpt: 0.168807 acc_pose: 0.291763
2025/11/11 07:19:33 - mmengine - INFO - Epoch(train) [210][ 850/1232] base_lr: 2.000822e-04 lr: 2.000822e-04 eta: 0:01:52 time: 0.321862 data_time: 0.097282 memory: 3062 loss: 0.168054 loss_kpt: 0.168054 acc_pose: 0.290739
2025/11/11 07:19:47 - mmengine - INFO - Epoch(train) [210][ 900/1232] base_lr: 2.000621e-04 lr: 2.000621e-04 eta: 0:01:37 time: 0.279979 data_time: 0.073039 memory: 3062 loss: 0.169768 loss_kpt: 0.169768 acc_pose: 0.346018
2025/11/11 07:20:01 - mmengine - INFO - Epoch(train) [210][ 950/1232] base_lr: 2.000449e-04 lr: 2.000449e-04 eta: 0:01:23 time: 0.265228 data_time: 0.060767 memory: 3062 loss: 0.173152 loss_kpt: 0.173152 acc_pose: 0.261617
2025/11/11 07:20:13 - mmengine - INFO - Epoch(train) [210][1000/1232] base_lr: 2.000304e-04 lr: 2.000304e-04 eta: 0:01:08 time: 0.258469 data_time: 0.059212 memory: 3062 loss: 0.169629 loss_kpt: 0.169629 acc_pose: 0.279666
2025/11/11 07:20:26 - mmengine - INFO - Epoch(train) [210][1050/1232] base_lr: 2.000188e-04 lr: 2.000188e-04 eta: 0:00:53 time: 0.244850 data_time: 0.053436 memory: 3062 loss: 0.175702 loss_kpt: 0.175702 acc_pose: 0.375336
2025/11/11 07:20:38 - mmengine - INFO - Epoch(train) [210][1100/1232] base_lr: 2.000099e-04 lr: 2.000099e-04 eta: 0:00:38 time: 0.252764 data_time: 0.058202 memory: 3062 loss: 0.174693 loss_kpt: 0.174693 acc_pose: 0.310100
2025/11/11 07:20:51 - mmengine - INFO - Epoch(train) [210][1150/1232] base_lr: 2.000039e-04 lr: 2.000039e-04 eta: 0:00:24 time: 0.246998 data_time: 0.052911 memory: 3062 loss: 0.171328 loss_kpt: 0.171328 acc_pose: 0.219182
2025/11/11 07:21:03 - mmengine - INFO - Epoch(train) [210][1200/1232] base_lr: 2.000006e-04 lr: 2.000006e-04 eta: 0:00:09 time: 0.251320 data_time: 0.057503 memory: 3062 loss: 0.165283 loss_kpt: 0.165283 acc_pose: 0.214718
2025/11/11 07:21:11 - mmengine - INFO - Exp name: rtmpose-m_8xb32-210e_coco-wholebody-hand-256x256_20251111_041351
2025/11/11 07:21:11 - mmengine - INFO - Saving checkpoint at 210 epochs
2025/11/11 07:21:18 - mmengine - INFO - Epoch(val) [210][50/62] eta: 0:00:00 time: 0.080297 data_time: 0.021689 memory: 3062
2025/11/11 07:21:19 - mmengine - INFO - Evaluating PCKAccuracy (normalized by "bbox_size")...
2025/11/11 07:21:19 - mmengine - INFO - Evaluating AUC...
2025/11/11 07:21:19 - mmengine - INFO - Evaluating EPE...
2025/11/11 07:21:19 - mmengine - INFO - Epoch(val) [210][62/62] PCK: 0.778149 AUC: 0.815162 EPE: 5.338884 data_time: 0.020232 time: 0.078186
Whether the loss or the accuracy, it is different from what you provided in rtmpose_coco_wholebody_hand.md, and the test results were very poor. what should I do ? please.
Reproduces the problem - command or script
tools/dist_train.sh configs/hand_2d_keypoint/rtmpose/coco_wholebody_hand/rtmpose-m_8xb32-210e_coco-wholebody-hand-256x256.py 1
Reproduces the problem - error message
no error ,just the result is not good.
Additional information
No response
Source: open-mmlab/mmpose