lelimite4444 / bridgedepthflow Goto Github PK

Hi,

Sorry to bother you again.

According to your paper, the 2-warp consistency loss should compare reconstructed $\ddot{I}$ via 2 warp and $\tilde{I}$ that reconstructed directly.

However, in your train.py code, take the module(a) for example,

if args.type_of_2warp == 1:
    mask_4 = [fw_mask[i][[2,3]] for i in range(4)]
    warp2_est_4 = [Resample2d()(left_est[i][[0,1]], disp_est_scale[i][[2,3]]) for i in range(4)]
    loss += 0.1 * sum([warp_2(warp2_est_4[i], left_pyramid[i][[6,7]], mask_4[i], args) for i in range(4)])
    mask_5 = [bw_mask[i][[2,3]] for i in range(4)]
    warp2_est_5 = [Resample2d()(left_est[i][[6,7]], disp_est_scale_2[i][[2,3]]) for i in range(4)]
    loss += 0.1 * sum([warp_2(warp2_est_5[i], left_pyramid[i][[0,1]], mask_5[i], args) for i in range(4)])

from my understanding, the warp2_est_4 corresponds to $\ddot{L_1}$ and warp2_est_5 corresponds to $\ddot{L_2}$, and you use left_pyramid[i][[6,7]] and left_pyramid[i][[0,1]] to compute warp_2 loss, which means you use the original L1 and L2, rather than the directly reconstructed results $\tilde{L_1}$ and $\tilde{L_2}$ (I think should be left_est[i][[6,7]] and left_est[i][[0,1]]?)here.

Is there any typo here, or do I have misunderstanding?

Looking forward to your reply. Appreciate your help!

File "BridgeDepthFlow-master/models/networks/correlation_package/correlation.py", line 4, in
import correlation_cuda
ImportError: libcudart.so.9.2: cannot open shared object file: No such file or directory

Hello, I tried this code with cuda-9.0 and cuda-10.0 with pytorch-1.1.0, and I have built and installed the correlation module. But when I run python train.py, it gets an error above, should I use cuda-9.2?

Hi!

Appreciate your code and work. I have small questions with the pre-trained model you provided.
Is the model monodepth_ver_a/b/c correspond to the full-1/2/3 in the Table 1 and Table 2 in your paper? I run your test and evaluation script with monodepth_ver_a on KITTI2015 and get the results like:
abs_rel, sq_rel, rms, log_rms, d1_all, a1, a2, a3
0.0648, 0.8067, 4.186, 0.157, 8.623, 0.946, 0.979, 0.990
which is slightly different from Table 1. So I'm not sure if the model is the corresponding one or I missed some details for evaluation.

Hello, thank you for your code. I have a question. When I finished testing kitti-2015,
I got the test result. So how do I get my depth map?
I want to get depth maps to compare with GT.

hi, I follow your guidance in README.md. I use kitti split, use kitti_train_files_png_4frames.txt for training, use kitti_stereo_2015_test_files.txt to generate disparities, use "/training/disp_occ_0/" as gt. 2warp type is 2. I don't change any parameters in your train.py except batchsize. But I get abs_rel=0.0763, sq_rel=0.8468, etc. Which data should I compare with in your Table1 in your essay?

Hi,
Excellent work! It is a good idea to combine the stereo matching and optical flow and use a single network. I have some questions. I saw the default epoch is set to 80. But, the time of training 80 epochs is very huge. How many epochs you have trained? How long does it take to train the entire model?
Best，
Zhai

Hi,

I can't seem to find your results on KITTI stereo or flow website under the name "BridgeDepthFlow". Is your work registered under a different name when submitted to KITTI?

Thank you