DLR-RM / AugmentedAutoencoder

Official Code: Implicit 3D Orientation Learning for 6D Object Detection from RGB Images
MIT License
338 stars 97 forks source link

Bug while loading ground truth masks #77

Closed safoex closed 4 years ago

safoex commented 4 years ago

Describe the bug Attempting to run evaluation with ground truth bboxes, I got an error with loading ground truth masks: Traceback (most recent call last): File "/home/safoex/.local/bin/ae_eval", line 8, in <module> sys.exit(main()) File "/home/safoex/Documents/libs/AugmentedAutoencoder/auto_pose/eval/ae_eval.py", line 136, in main test_img_crops, test_img_depth_crops, bbs, bb_scores, visibilities = eval_utils.get_gt_scene_crops(scene_id, eval_args, train_args, load_gt_masks = gt_masks) File "/home/safoex/Documents/libs/AugmentedAutoencoder/auto_pose/eval/eval_utils.py", line 66, in get_gt_scene_crops (H_AE, W_AE), visib_gt=visib_gt,inst_masks=gt_inst_masks) File "/home/safoex/Documents/libs/AugmentedAutoencoder/auto_pose/eval/eval_utils.py", line 195, in generate_scene_crops mask = inst_masks[view] IndexError: list index out of range Most likely, this is because code references local path e.g. here.

System Info Describe the characteristic of your environment:

Additional context There were few other small bugs, mostly with py2-3 compatibility in provided replacement for sixd_toolkit. E.g. wrong tabulation here and next line. As soon as I manage to run tests correctly, I will be happy to make a pull request resolving everything I faced!

MartinSmeyer commented 4 years ago

Thanks for testing @safoex

Sorry about the local path, I removed it. I adapted the README, you can now download the gt_masks here and point the 'external' parameter in the eval config to that path.

MartinSmeyer commented 4 years ago

PS: PRs are always welcome, thanks!

0smile commented 1 year ago

@safoex Hi, thanks for your testing. I wonder that how do you successfully train the model on gtx1650? I modified the batch-size to 1 but still raise "Resource exhausted" error. Looking for your replying!