zyf12389 / gc-net Goto Github PK

Hello,

I am almost a freshman. May I ask you something？I download the Scene Flow Dataset，but the dataSet I downloaded is in png format. How should I use it in the project? Could you give me a hint where you acquired these data?

Thank you.

Hi, it would be nice, if possible, to provide the trained weights along with the model for reproducibility.

Hello,

I am almost a freshman in computer vision so my question might be not so brilliant.

In the scripts main.py and read_data.py, files of formats '.t7' and '.pkl' should be loaded. However, I did not find any file with these suffixes in the Scene Flow Dataset. Could you give me a hint where you acquired these data?

Thank you.

Hi, Thank you for releasing the code.
I have tried your code, but only trained the flyingthing3D dataset.
It all looks well on training and testing on flyingthing3D.
However, when I tried on my own dataset which is similar to flyingthing, the result is a total mess up. I mean that even can't be recognized as a disparity map.

This is my own data

Below is my own data's GT and disp

And it is the test data in flyingthing(GT and disp)

I can't figure out what's wrong with my result.
Is it possible that I only train for 5 epoch? or something other like lighting ?
BTW, is it possible to release your pretrained model to let me check with it?

Thank you

Hi, @zyf12389
Thanks for you code! Could you tell me that can this implementation achieves the result authors shown in GC-NET paper or how this implementation performs?

In the paper it mentions that it repeats steps 2 and 3 in Table 1, 7 times thereafter. Am I missing something in the implementation that doesn't include this seven more times?