-
Notifications
You must be signed in to change notification settings - Fork 4
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
I have some questions about training on VOC datasets #23
Comments
In case of VOC, follow this config file. Both should be seen. It has been a while since I uploaded the codes, and I don't exactly remember why we kept them like this. Maybe we have hardcoded something in some intermediate file. But it is correct. |
When to set unsee categories,test? |
Are the settings the same for ZSD and GZSD? |
In case of VOC especially, just keep it exactly the same as mentioned in this file. |
For VOC dataset, ZSD/GZSD setting was ill-defined when we did this work. If you look at our paper, for VOC, we have just shown seen and unseen class results - nothing explicitly defined as ZSD/GZSD as in MSCOCO. To summarize, for VOC we just did one experiment following previous works - just find the results separately on seen and unseen classes and report them. And that experiment shall be done following our command in step 6, replacing with voc related weight files and config files. Please use the "zsd" flag for evaluation on VOC. |
During the training and testing processes of a generative model(The fifth step and the sixth step). Do both the
val
andtest
sets in the VOC dataset configs file need to be set to 'seen' or unseen ?val=dict(
type=dataset_type,
ann_file=data_root + 'VOC2007/ImageSets/Main/val.txt',
img_prefix=data_root + 'VOC2007/',
classes_to_load='seen',
split=split,
pipeline=test_pipeline),
test=dict(
type=dataset_type,
ann_file=data_root + 'VOC2007/ImageSets/Main/test.txt',
img_prefix=data_root + 'VOC2007/',
split=split,
classes_to_load='seen',
pipeline=test_pipeline))
The text was updated successfully, but these errors were encountered: