kinglittleq / psmnet Goto Github PK

Thanks very much for providing the pytorch-reimplementation of PSMnet.

Do you reproduce the result of original paper or match the result of the original code? Can you provide the evaluation result for KITTI dataset ?

Good evening,
I'm running the inference.py using your best pretrained model (best_model.ckpt ) using the following images from Kitti_2015 testing set
Left img

Right img

And i'm getting the following disparity map, very blurry as you can see... am i missing something?
Thanks in advance!

作者你好，训练的时候报错说GPU空间不足，请问你训练用的配置是什么？

PyTorch 1.0+ and CUDA 10 support would be a useful.

Hello
When I run to the verification step, the 3px-error rate always shows 0. Can you tell me where there is a problem？
I found that target_disp read the value of 0, but, I don't know where it is.
Can you help me with it？

Hi, how much time do you need to finish an epoch?
I try to train the module one the 4 1080ti GPU, but it looks like stuck at the begin of the model.
it even do not finish an epoch..

hi there:
I changed the batchsize to 1, and run the code.
There is an
RuntimeError: CUDA out of memory. Tried to allocate 352.00 MiB (GPU 0; 5.93 GiB total capacity; 3.99 GiB already allocated; 32.56 MiB free; 782.19 MiB cached).

What's the GPU you used? and how many time you used to train the model?

self.disp_score = torch.range(0, max_disp - 1) # [D]
Traceback (most recent call last):
File "train.py", line 215, in
main(args)
File "train.py", line 75, in main
model.load_state_dict(state['state_dict'])
File "/home/cb/anaconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 719, in load_state_dict
self.class.name, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for PSMNet:
Missing key(s) in state_dict: "cost_net.cnn.conv0.0.net.0.weight", "cost_net.cnn.conv0.0.net.1.weight", "cost_net.cnn.conv0.0.net.1.bias", "cost_net.cnn.conv0.0.net.1.running_mean", "cost_net.cnn.conv0.0.net.1.running_var", "cost_net.cnn.conv0.1.net.0.weight", "cost_net.cnn.conv0.1.net.1.weight",