Giters

Raykoooo / IAST

IAST: Instance Adaptive Self-training for Unsupervised Domain Adaptation (ECCV 2020)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Unable to use pretrained models

miguel-dgist opened this issue · comments

commented

Hi, I've been unsuccessfully trying to reproduce your results. Neither train nor test seem to be working on my machine, all the results are almost zero. Here is what I got when I tried testing:

$ python eval.py --config_file config/gtav2cityscapes_t4/run_task/sl_3.yaml --resume_from pretrained_models/gta5_t4_best.pth 
Resume from: pretrained_models/gta5_t4_best.pth
Eval Size: [[1024, 512], [1280, 640], [1536, 768], [1800, 900], [2048, 1024]]
Use Flip: False
100%|█████████████████████████████████████████| 250/250 [11:54<00:00,  2.86s/it]
val_miou: 0.0049, 0: 0.0000, 1: 0.0001, 2: 0.0117, 3: 0.0266, 4: 0.0358, 5: 0.0062, 6: 0.0000, 7: 0.0000, 8: 0.0001, 9: 0.0054, 10: 0.0000, 11: 0.0020, 12: 0.0008, 13: 0.0008, 14: 0.0000, 15: 0.0003, 16: 0.0000, 17: 0.0029, 18: 0.0000

I will appreciate your suggestions!

commented

Hi, This seems to be an error in the label you read. If you have not modified any Dataloader related code, please check whether your cityscapes image and label are correct.

commented

Thank you for your answer. Actually the only issue I had was that the name of cityscapes labels has a "Train" word added to the original ones I downloaded from cityscapes website.

The file "IAST/data/cityscapes_train.json" has:
"mask_name": "gtFine/train/aachen/aachen_000000_000019_gtFine_labelTrainIds.png"

while the data I got came with the name:
"gtFine/train/aachen/aachen_000000_000019_gtFine_labelIds.png"

Is there any preprocessing step that actually modifies these labels?

commented

The file "IAST/data/cityscapes_train.json" has:
"mask_name": "gtFine/train/aachen/aachen_000000_000019_gtFine_labelTrainIds.png"

while the data I got came with the name:
"gtFine/train/aachen/aachen_000000_000019_gtFine_labelIds.png"

Is there any preprocessing step that actually modifies these labels?

Please refer to https://github.com/mcordts/cityscapesScripts.

commented

It's working now. Thank you!

commented

Hi, I am facing the same issue. I generated the labelTrainIds images using createTrainIdLabelImgs.py. However, it still gives the same issue (mIoU ~0.009). Could you please let me know how exactly you preprocess the data? Thanks.

commented

Hi @akshaykvnit, I did not perform any additional preprocessing steps but I had only check the Tesla T4 pretrained model.

commented

Hi, thank you for releasing code and checkpoints. However, I am not able to reproduce the results in the repository table (51.88 and 51.2) for GTA. Are these results obtained with multi-scaling? if not what resolution do you use? Other works typically use 1024x512 at test time, but this does not seem to work.

Thank you in advance,