Enze Xie comments

Results 36 comments of


                                            Enze Xie

Code crashes for new version of TF and Keras

i have the same problem

The results seems not good.

I train this model on Street View dataset,my result doesn't good as well. Do anybody achieve good result?

crash after recent commits: assert manifold_w * manifold_h == num_images

hi , I meet this problem also,do you find way to solve this problem?

crash after recent commits: assert manifold_w * manifold_h == num_images

ha I find it! the batch_size must be set as n * n,such as 16(4 * 4),64(8*8,default)or other. otherwise the assert will throws the exception. but I don't know why.

训练SegFormer-B1在cityscapes数据集，出现错误“Defaule process group is not initialized”

please use 'dist_train.sh' instead of 'train.py', because I use 'syncbn' in the model. It requires pytorch ddp.

Porting SegFormer to HuggingFace Transformers

Hi, Thanks! It is so great if SegFormer will be in HuggingFace! For your questions: > (0) basic image + mask transformations for inference and finetune on custom datasets First,...

Porting SegFormer to HuggingFace Transformers

Hi, Although `random flip` is defined in the config's `test_pipeline`, it is not used for inference unless you set `aug-test=True`(means multi-scale+flip test) in `tools/test.py` to evaluate the dataset. But for...

Porting SegFormer to HuggingFace Transformers

Hi, The layer `conv_seg` is not used, you can remove it. The documentation page looks very nice!

Porting SegFormer to HuggingFace Transformers

Hi, I am not sure whether use CV2 or PIL to read images will cause a slight difference. I think you can visualize the result and compare them. Also, you...

Porting SegFormer to HuggingFace Transformers

Hi @NielsRogge I have finished the vis code in colab, please check it. If you do not want to involve `matplotlib`, you can save the image using PIL instead of...