search

a1600012888 / MUTR3D

Code for Paper, MUTR3D: A Multi-camera Tracking Framework via 3D-to-2D Queries. https://tsinghua-mars-lab.github.io/mutr3d/
MIT License
179 stars 19 forks source link

error occurred and there's no results update after 40 epoch as shown in this command after eval #7

Closed XuanjieXiao closed 2 years ago

XuanjieXiao commented 2 years ago

2022-07-21 09:04:35,386 - mmdet - INFO - Epoch(val) [44][321] pts_bbox_NuScenes/amota: 0.0000, pts_bbox_NuScenes/amotp: 2.0000, pts_bbox_NuScenes/recall: 0.0000, pts_bbox_NuScenes/motar: 0.0000, pts_bbox_NuScenes/gt: 611.5000, pts_bbox_NuScenes/mota: 0.0000, pts_bbox_NuScenes/motp: 2.0000, pts_bbox_NuScenes/mt: 0.0000, pts_bbox_NuScenes/ml: 203.0000, pts_bbox_NuScenes/faf: 500.0000, pts_bbox_NuScenes/tp: 0.0000, pts_bbox_NuScenes/fp: 0.0000, pts_bbox_NuScenes/fn: 3669.0000, pts_bbox_NuScenes/ids: 0.0000, pts_bbox_NuScenes/frag: 0.0000, pts_bbox_NuScenes/tid: 20.0000, pts_bbox_NuScenes/lgd: 20.0000

sorry to bother you again ! I've trained this for a long time! Suddenly, I saw this results in the training log. So I deploy the eval command like you told us to do. But I got this results and this error. I've looked all over the Internet , but I cannot find a solution. image

AMOTA 0.000 AMOTP 2.000 RECALL 0.000 MOTAR 0.000 GT 14556 MOTA 0.000 MOTP 2.000 MT 0 ML 6470 FAF 500.0 TP 0 FP 0 FN 101897 IDS 0 FRAG 0 TID 20.00 LGD 20.00

what's more, my command is listed below. python tools/test.py Test01/Full_Train/resnet101_fpn_3frame_full.py Test01/Test02/epoch_45.pth --eval bbox

But when I use the pretrained model like "mutr3d_res101.pth", which is downloaded from your GitHub, the program can run correctly. looking forward to your reply ! thks

a1600012888 commented 2 years ago

Hi, I suspect your model went to nan. Can you generate the results json file, python3 tools/test.py plugin/track/configs/resnet101_fpn_3frame.py <model-path> --format-only --eval-options jsonfile_prefix=<dir-name-for-saving-json-results> I believe the model will only output nan for any input.

XuanjieXiao commented 2 years ago

Hi, I suspect your model went to nan. Can you generate the results json file, python3 tools/test.py plugin/track/configs/resnet101_fpn_3frame.py <model-path> --format-only --eval-options jsonfile_prefix=<dir-name-for-saving-json-results> I believe the model will only output nan for any input.

yes! I can generate the results json file. results_nusc.zip

this is the results that I use the pretrained model image

image

ghost commented 1 year ago

2022-07-21 09:04:35,386 - mmdet - INFO - Epoch(val) [44][321] pts_bbox_NuScenes/amota: 0.0000, pts_bbox_NuScenes/amotp: 2.0000, pts_bbox_NuScenes/recall: 0.0000, pts_bbox_NuScenes/motar: 0.0000, pts_bbox_NuScenes/gt: 611.5000, pts_bbox_NuScenes/mota: 0.0000, pts_bbox_NuScenes/motp: 2.0000, pts_bbox_NuScenes/mt: 0.0000, pts_bbox_NuScenes/ml: 203.0000, pts_bbox_NuScenes/faf: 500.0000, pts_bbox_NuScenes/tp: 0.0000, pts_bbox_NuScenes/fp: 0.0000, pts_bbox_NuScenes/fn: 3669.0000, pts_bbox_NuScenes/ids: 0.0000, pts_bbox_NuScenes/frag: 0.0000, pts_bbox_NuScenes/tid: 20.0000, pts_bbox_NuScenes/lgd: 20.0000

sorry to bother you again ! I've trained this for a long time! Suddenly, I saw this results in the training log. So I deploy the eval command like you told us to do. But I got this results and this error. I've looked all over the Internet , but I cannot find a solution. image

AMOTA 0.000 AMOTP 2.000 RECALL 0.000 MOTAR 0.000 GT 14556 MOTA 0.000 MOTP 2.000 MT 0 ML 6470 FAF 500.0 TP 0 FP 0 FN 101897 IDS 0 FRAG 0 TID 20.00 LGD 20.00

what's more, my command is listed below. python tools/test.py Test01/Full_Train/resnet101_fpn_3frame_full.py Test01/Test02/epoch_45.pth --eval bbox

But when I use the pretrained model like "mutr3d_res101.pth", which is downloaded from your GitHub, the program can run correctly. looking forward to your reply ! thks

Hi, I meet the same problem with you, how do you solve it later ?

a1600012888 commented 1 year ago

Hi Qingfeng.

Have you tested if your model output nan?