deep-high-resolution-net.pytorch

deep-high-resolution-net.pytorch copied to clipboard

AI Challenger trained HRNet weights

4

Hi, thanks for your great work. At `README.md`, there's an evaluation result of HRNet trained on MSCOCO+__AI Challenger__. (Indicated with *) Where can I found the pretrained weights of that...

hongsukchoi

Dear all, HRNet is really a good network. Currently, I am training a HRNet model to predict key point for my own 3D data. However, overfitting happened in my training...

XiaotianJia

Hi @leoxiaobin , thanks for your great work: HRNet. As readme files mentioned: pose_hrnet_w48* means using additional data from AI challenger for training. As we all know, AI Challenge has...

dzyjjpy

CUDA out of memory

3

### How to fix it? ### The error messages are below: Number of Layers Conv2d : 293 layers BatchNorm2d : 292 layers ReLU : 261 layers Bottleneck : 4 layers...

s692833

ERROR using numpy with version >=1.18

1

an error appeared when I use numpy 1.18 for this code: Traceback (most recent call last): File "tools/train.py", line 210, in main() File "tools/train.py", line 180, in main final_output_dir, tb_log_dir...

guihuzhang

Generally, how can I modify the structure of the network? Like changing layers, adding transformation, and so on.

Tabriz12

'center' and 'scale' in MPII annotation

3

Hi! How can I get the 'center' and 'scale' in MPII annotation? Are these values obtained by a detector? Or they're just the average of the joints coordinates?

violetnotfat

Hi, I think that there is an error in the calculus of the PCK accuracy, as reported [here](https://github.com/stefanopini/simple-HRNet/issues/66) and [here](https://github.com/microsoft/human-pose-estimation.pytorch/issues/144). At this line https://github.com/leoxiaobin/deep-high-resolution-net.pytorch/blob/ba50a82dce412df97f088c572d86d7977753bf74/lib/core/evaluate.py#L55 `[h, w]` shouldn't be inverted to...

stefanopini

how can i train on my data

1

if i want train on my data how should i change gt_valid.mat

asmallcodedog

Why do the head and neck locations of some instances in train.json and val.json not integer-valued? Taking the first sample in train.json as an example, the 9th joint is at...

fmu2