SpadeLiu
SpadeLiu
Hi, Did you use "train.py" to train a model on SceneFlow? They share the same code.
> Have you conducted multiple experiments and have the results of different epochs been stable? Hi, I didn't explicitly do this but when I search the optimal hyper-parameters the results...
> hi, congratulations! Nice work! I have a question about /networks/deformable_refine.py/class GetValueV2(nn.Module)/line 72 : > > ``` > # clip p > p_y = torch.clamp(p[..., :N], 0, h-1) / (h-1)...
Hi， Do you mean the method that render the disparity maps to color maps? The toolkit is from KITTI benchmark [http://www.cvlibs.net/datasets/kitti/eval_scene_flow.php?benchmark=stereo](url). It is a Matlab toolkit.
Yes, sorry for the error.
> @SpadeLiu Have you tested network backbone such as ConvNext or Transformer? Thanks for your suggestion. We have not tested these models due to the lack resources. But I think...
> @SpadeLiu Thanks for your reply, I found that using deep features of these models does reduce generalization ability and they do not perform as well as simpler models like...
@xzjzsa I do not think the batch-size affects the performance a lot in this code. When training the adaptor, can you discard the udc loss and only use the smooth...
@xzjzsa Yes. It seems when the cost aggregation module is fixed, the udc loss will not guide the learning process well.
@xzjzsa I can not explain this. In my opinion, if the cost volume is built by cosine similairy, the performance variation will not be so large.