search

zhengye1995 / Zero-shot-Instance-Segmentation

code for CVPR paper Zero-shot Instance Segmentation
Apache License 2.0
93 stars 14 forks source link

Is train dataset config is correct? #11

Closed frkake closed 3 years ago

frkake commented 3 years ago

Below train config setting is correct? It looks ignoring seen classes even during training. I wonder why this data class is called during training.

https://github.com/zhengye1995/Zero-shot-Instance-Segmentation/blob/6b25b318d7821b5f7a48f7d5e6dbd0bc63474e14/configs/zsi/65_15/train/zero-shot-mask-rcnn-BARPN-bbox_mask_sync_bg_65_15_decoder_notanh.py#L141

https://github.com/zhengye1995/Zero-shot-Instance-Segmentation/blob/6b25b318d7821b5f7a48f7d5e6dbd0bc63474e14/configs/zsi/48_17/train/zero-shot-mask-rcnn-BARPN-bbox_mask_sync_bg_decoder.py#L142

I think that CocoDataset_48_17 and CocoDataset_65_15 data class looks appropriate.

Thank you a lot!

zhengye1995 commented 3 years ago

Sorry for this confusing code. In fact, it only affects the category name during visualization.The data and categories used for training are completely determined by the label file https://github.com/zhengye1995/Zero-shot-Instance-Segmentation/blob/6b25b318d7821b5f7a48f7d5e6dbd0bc63474e14/configs/zsi/48_17/train/zero-shot-mask-rcnn-BARPN-bbox_mask_sync_bg_decoder.py#L176. (note: 'dataset_type' affects training in the new version of mmdetection, but has no effect in this old version)

frkake commented 3 years ago

@zhengye1995 Thank you so much for fast response! I'm convinced very well!

By the way, what version of mmdetection is based in this repository? I would like to develop this in upper version of it.

zhengye1995 commented 3 years ago

mmcv 0.2.14
mmdet 1.0rc0+45ee140

frkake commented 3 years ago

@taku-buntu Thank you so much!