bbox loss is 0.0000

[kkissmart]

Hi, I am trying to train the model based on your config,

but my bbox loss is 0.00000

I'm doing the first step of table detection

logs are pasted as below

index created! 2020-08-19 04:52:26,757 - mmdet - INFO - Start running, host: yinhan@thinkmate, work_dir: /home/yinhan/mmdetection/work_dirs/cascade_mask_rcnn_hrnetv2p_w32_20e 2020-08-19 04:52:26,757 - mmdet - INFO - workflow: [('train', 1)], max: 50 epochs /home/yinhan/miniconda3/envs/cv/lib/python3.8/site-packages/torch/nn/functional.py:2970: UserWarning: Default upsampling behavior when mode=bilinear is changed to align_corners=False since 0.4.0. Please specify align_corners=True if the old behavior is desired. See the documentation of nn.Upsample for details. warnings.warn("Default upsampling behavior when mode={} is changed " /home/yinhan/miniconda3/envs/cv/lib/python3.8/site-packages/torch/nn/functional.py:2970: UserWarning: Default upsampling behavior when mode=bilinear is changed to align_corners=False since 0.4.0. Please specify align_corners=True if the old behavior is desired. See the documentation of nn.Upsample for details. warnings.warn("Default upsampling behavior when mode={} is changed " 2020-08-19 04:53:07,442 - mmdet - INFO - Epoch [1][50/70130] lr: 4.784e-04, eta: 32 days, 23:22:04, time: 0.812, data_time: 0.252, memory: 5684, loss_rpn_cls: 0.2360, loss_rpn_bbox: 0.0000, s0.loss_cls: 0.6219, s0.acc: 96.7568, s0.loss_bbox: 0.0000, s1.loss_cls: 0.4192, s1.acc: 92.0176, s1.loss_bbox: 0.0000, s2.loss_cls: 0.2595, s2.acc: 86.6953, s2.loss_bbox: 0.0000, loss: 1.5365

which part is wrong?

the data (tablebank) looks like: {'segmentation': [[85.0, 396.0, 85.0, 495.0, 510.0, 495.0, 510.0, 396.0]], 'area': 42075.0, 'image_id': 1, 'category_id': 1, 'id': 1, 'iscrowd': 0, 'bbox': [85.0, 396.0, 425.0, 99.0]}

thanks so much!

[kkissmart]

I forgot to reset the classes for dataset.

My bad