Log images during mmdetection training

W&B Help
1

I am using the mmdetection platform.
For training I am using a .py config file that contains the log configuration from W&B documentation:
log_config = dict(
hooks = [
dict(type=‘TextLoggerHook’),
dict(type=‘MMDetWandbHook’,
init_kwargs={‘project’: ‘…’, ‘entity’: ‘…’, ‘name’: ‘…’},
interval=10,
log_checkpoint=True,
log_checkpoint_metadata=True,
num_eval_images=200,
bbox_score_thr=0.3)]
)
For some reason, I can’t log images of the train, validation, and their ground truth and predictions during training.
Did someone do it and can give me a good tip for that?
Thanks,
Ziv

2

Hi @zivpumba, could you send me a link to the run in the UI?

Also, after you run:

config_file = 'mmdetection/configs/path/to/config.py'
cfg = Config.fromfile(config_file)

can you verify that:

cfg.log_config.hooks

Returns the hooks list correctly?

Thank you,
Nate

3

Hi Nate,
This is how my cfg.log_config.hooks looks like:

[{‘type’: ‘TextLoggerHook’}, {‘type’: ‘MMDetWandbHook’, ‘init_kwargs’: {‘project’: ‘Cars_Detector’, ‘entity’: ‘pumba’, ‘name’: ‘exp23’}, ‘interval’: 10, ‘log_checkpoint’: True, ‘log_checkpoint_metadata’: True, ‘num_eval_images’: 200, ‘bbox_score_thr’: 0.3}]

4

@zivpumba, sorry for the delay on this. Could you send me a link to your run? The image table is likely being logged but is available in the Artifacts tab and isn’t getting sent to your workspace by default.

Thank you,
Nate

5

Hi @zivpumba, I wanted to follow up on this and see if you still needed help with this? If so, could you send a link to your workspace and I can take a look?

Thank you,
Nate

6

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.