No visualization, logs and snapshots

[BobaBobek]

Hi Thaoshibe, I try to train network by:

python train.py --data_path /path/to/dataset --epochs=5000 --vis_step=100 --task_name=ganmakeup002

I already trained 368 epochs from 5 000 but I am not able to find any output in these subdirectories /vizualization/_ganmakeup002 , /snapshot/_ganmakeup002, /logs/_ganmakeup002. I expected some outputs every 100 epochs (--vis_step=100). Did I miss some parameter?

[thaoshibe]

Hi there, I've tested the code and it ran smoothly in my local machine. In the visualization folder, you should find something like this

I'm sorry to say that I can not help you without further details.

Please try to debug from here

Have a good day! (.❛ ᴗ ❛.)

[BobaBobek]

Hi, I found out that there is a difference between trainig with standard Makup Transfer Dataset and training with Makup Transfer Dataset with UV-map texture (from project VinAIResearch/CPM). The logging of training with standard Makup Transfer Dataset is working fine, but the logging of training with Makup Transfer Dataset with UV-map texture is not working, because this condition is TRUE every time and logging is skiped:

                    # avoid the situation that images with eye closed
                    if not ((mask_A_eye_left>0).any() and (mask_B_eye_left>0).any() and \
                        (mask_A_eye_right > 0).any() and (mask_B_eye_right > 0).any()):
                        continue

Any idea how to overcome this for Makup Transfer Dataset with UV-map texture? Removing this condition is not working.

[thaoshibe]

I can not come up with any reason for that. Let's say, just comment out those lines and see what'll happen?

[BobaBobek]

When I comment this piece of code, it crashes inside rebound_box function on line 213: mask_A_face[: ,: ,min(x_A_index)-10:max(x_A_index)+11, min(y_A_index)-10:max(y_A_index)+11] ValueError: min() arg is an empty sequence

[thaoshibe]

Then I'd suggest you to double-check your mask 🤔 It should be facial segmentation and have value as mentioned here

[thaoshibe]

Hello, have you solve the issue? :<

[BobaBobek]

-Unfortunately no :-(