Peize Sun

Hi~ For training the model from scratch, we didn't try it. I guess it may need more training epochs. For the init boxes, please see this issue: [link](https://github.com/PeizeSun/SparseR-CNN/issues/47).

```python self.init_proposal_boxes = nn.Embedding(self.num_proposals, 4) if self.init_type == "img": nn.init.constant_(self.init_proposals.weight[:, :2], 0.5) nn.init.constant_(self.init_proposals.weight[:, 2:], 1.0) elif self.init_type == "center": nn.init.constant_(self.init_proposals.weight[:, :2], 0.5) nn.init.constant_(self.init_proposals.weight[:, 2:], 0.1) elif self.init_type == "grid": scale...

@liz6688 If you find any bug when using different initializations, please remind me, thanks~

Hi~ Our code is based on Detectron2, the way of using custom dataset is the same as detectron2, I guess you could refer to [Use Custom Dataset](https://detectron2.readthedocs.io/en/latest/tutorials/datasets.html).

Hi~ Please refer to [issues 47](https://github.com/PeizeSun/SparseR-CNN/issues/47)

Hi~ The default TTA code in detectron2 doesn't work for sparse r-cnn. We have uploaded the right TTA code. You can use it by ``` python projects/SparseRCNN/train_net.py --num-gpus 8 --config-file...

Hi~ It seems you are installing on Windows OS. Unluckily, our code doesn't support Windows OS.

Hi~ We tried detach both, keep both, detach one and keep another, current version is the setting with best result.

> It seems that self.decoder_track is trained under torch.randn(1).item() > 0.0 condition, self.decoder is trained otherwise. There are two questions: > > 1. when we train self.decoder, the two part...

Hi~ For multiclass object tracking, you should change models/tracker.py