TypeError: forward_train() got an unexpected keyword argument 'gt_bboxes_ignore

[qilicun]

Hi,

Firstly, thank you very much for this work, great job! Secondly, when I tried to run this detection follow the introduction with VOC dataset, there was an error TypeError: forward_train() got an unexpected keyword argument 'gt_bboxes_ignore I modified cfg_retinanet_ghm_r50_fpn_1x.py, replace the data config part with that in GHM_Detection/mmdetection/configs/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712.py, what did I miss?

Thirdly, do you have any plan to release your pretrained model? Thanks again!

[libuyu]

Thank you for your affirmation.

It seems you still keep some configs of COCO, because the ignore region is used in COCO with the crowd label but not in VOC.

If you still have no clue, you can send an email to me with your config file and tell me which line the error locates. But since I am attending the AAAI conference now, maybe I will reply you a few days later.

[qilicun]

@libuyu Thanks for your reply. After I comment the setting with_crowd=False in train and val dict, it finally works!

Do you have any suggestion or tutorial that if I use the prtrained model of GHM with COCO datasets to train my own data?

[libuyu]

Sorry, I forgot that. I just use the official model. https://download.pytorch.org/models/resnet50-19c8e357.pth

[libuyu]

The model is trained on image-net and in theory, it can be used for fine-tuning on any data set.

[qilicun]

To be specific, my own detection data contains less than 1000 images, I wonder if trained with the pre-trained model on COCO would be better than pre-trained with the IMAGE-NET one?

[libuyu]

I understand your concern. But I think both COCO and ImageNet are very large compared with your set. So maybe the pre-trained model is not the key. Although you can study the effect of different pre-trained models. But in my opinion, how to avoid overfitting in such a small set is more important.