ContrastiveSeg

ContrastiveSeg copied to clipboard

Hi! I want to reproduce the HRNet result trained on cityscapes trainval, could you please release related code?

ICE-Bro

Running the t code shows loss=nan when calculating the coco dataset.

4

Running the t code shows loss=nan when calculating the coco dataset.

liaochuanlin

@tfzhou So, the class 0 is the ignore index and not the background class? I mean, for the background class is applied also the contrastive loss. Right?

4

@tfzhou So, the class 0 is the ignore index and not the background class? I mean, for the background class is applied also the contrastive loss. Right? _Originally posted by...

zhangyuxuan321

Example for max_views, max_samples?

2

First of all, thanks a lot for you brilliant work. I really appreciate it. I have one quick question and would be glad if I could get an answer. How...

lennart-maack

Hi, I was checking the logging file, i.e., hrnet_w48_contrast_lr1x_hrnet_contrast_t0.1.log. The epoch and iteration seem to be computed as the training is for a single gpu, while it is a 4...

zhd2rng

Code bug

1

In trainer_contrastive.py line129 feat = this_feat[:, perm[:K]] should be idxs_perm = idxs[perm] feat = this_feat[:,idxs_perm[:K]].squeeze() Otherwise, it can not select the current points.

eraserNut

Semi-Hard Example Sampling implementation

3

Hello, I would like to ask if there is available the implementation of sampling hard positive/negatives pixel for the computation of the contrastive loss, because I only found the implementation...

nysp78

About the dimension of feature embedding

1

Hello, I would like to ask about the feature embedding from the backbone. Why not keep the same dimension as label through upsample.

chen-wolfberry

你好，loss_contrast与loss_contrast_mem两个版本中的PixelContrastLoss有哪些区别？是不是loss_contrast版本是采样了所有像素？还是随机采样的一个mini-batch的部分像素？

IceHowe

Hi... I am getting following error while resuming training from a checkpoint on a single GPU system. The training went fine when started from 0th iteration, but exited immediately after...

mailtohrishi