Error at Training the re-identification model. AssertionError: [!] Dataset not found: None

[ducmanhnguyen]

Hi,
At the step Training the re-identification model we ran:

python experiments/scripts/train_reid.py

After finish load the Train set, when it start to load the Val, the error came out as bellow:

INFO - main - [*] Val:
ERROR - train_reid - Failed after 0:29:46!
Traceback (most recent calls WITHOUT Sacred internals):
File "experiments/scripts/train_reid.py", line 82, in main
db_val = Datasets(db_val['split'], dataset_kwargs)
File "/root/tracking_wo_bnw/src/tracktor/datasets/factory.py", line 72, in init
assert dataset in _sets, f"[!] Dataset not found: {dataset}"
AssertionError: [!] Dataset not found: None

After check the config file provided at experiments/cfgs/reid.yaml I found out the the split of db_val is set as None . But I suppose it should be some valid name here.

After read through the code of factor.py , the reid set are defined from line 37-39, and only defined for train set, which not include the None:

tracking_wo_bnw/src/tracktor/datasets/factory.py

Lines 37 to 39 in cb5aaeb

	for split in ['train', '02', '04', '05', '09', '10', '11','13',]:
	name = f'mot_reid_{split}'
	_sets[name] = (lambda *args, split=split: MOTreIDWrapper(split, *args))

Please help us check what happened!

Thank you very much!

[tjusyy]

Hi,
At the step Training the re-identification model we ran:

python experiments/scripts/train_reid.py

After finish load the Train set, when it start to load the Val, the error came out as bellow:

INFO - main - [*] Val:
ERROR - train_reid - Failed after 0:29:46!
Traceback (most recent calls WITHOUT Sacred internals):
File "experiments/scripts/train_reid.py", line 82, in main
db_val = Datasets(db_val['split'], dataset_kwargs)
File "/root/tracking_wo_bnw/src/tracktor/datasets/factory.py", line 72, in init
assert dataset in _sets, f"[!] Dataset not found: {dataset}"
AssertionError: [!] Dataset not found: None

After check the config file provided at experiments/cfgs/reid.yaml I found out the the split of db_val is set as None . But I suppose it should be some valid name here.

After read through the code of factor.py , the reid set are defined from line 37-39, and only defined for train set, which not include the None:

tracking_wo_bnw/src/tracktor/datasets/factory.py

Lines 37 to 39 in cb5aaeb

	for split in ['train', '02', '04', '05', '09', '10', '11','13',]:
	name = f'mot_reid_{split}'
	_sets[name] = (lambda *args, split=split: MOTreIDWrapper(split, *args))

Please help us check what happened!

Thank you very much!

Hello, I have the same problem as you.
could you please tell me that have you solved it?

[timmeinhardt]

The MOT17 dataset has no validation set. So when training on the entire training set you can not validate. For the code to work just set:

db_val:
  split: mot_reid_train

Edit: I just pushed the newest version which works with a new reid module based on torchreid.

[ducmanhnguyen]

Thank you so much for your reply!

So basically, put the val set name as mot_reid_train is just like a trick to pass the error, and the training code will not run the validation at all, right? so I could set it as mot_reid_02, mot_reid_04 , etc. as long as it in the sets to not trigger the assertion error. is it true?

[timmeinhardt]

I should run the validation. But you are validating the same thing as you are training on.