Hi,
Good job!
I used your code to train the network, but found that the network converges too slow. I got "Average train loss = Stage 0 = 7.32 Stage 1 = 6.86 Stage 2 = 6.38 Stage 3 = 5.62" after 8 epoch. Is it normal? And could you please release the trained model?
Besides, you compared your network with StereoNet. I tried to re-implement StereoNet using PyTorch, but the runtime is much longer than that reported in the paper. I got about 0.17s when testing images from Scene Flow dataset on 1080 ti GPU. I wonder if you can achieve the runtime reported in StereoNet using your re-implementation. I will be very grateful if you could share the implementation of StereoNet too.
Thank you very much.
Hi, Good job! I used your code to train the network, but found that the network converges too slow. I got "Average train loss = Stage 0 = 7.32 Stage 1 = 6.86 Stage 2 = 6.38 Stage 3 = 5.62" after 8 epoch. Is it normal? And could you please release the trained model? Besides, you compared your network with StereoNet. I tried to re-implement StereoNet using PyTorch, but the runtime is much longer than that reported in the paper. I got about 0.17s when testing images from Scene Flow dataset on 1080 ti GPU. I wonder if you can achieve the runtime reported in StereoNet using your re-implementation. I will be very grateful if you could share the implementation of StereoNet too. Thank you very much.