Closed jamesjjcondon closed 3 years ago
Hi @jamesjjcondon ,
What's the PyTorch version that you are using? Here is the results for 0_R-MLO.png I obtained with PyTorch 1.1.0:
It seems to me that the patch selection logic does not behave as it's expected in your environment. For instance, in your 0_R-MLO.png, while the benign lesion in the lower breast is highlighted by the saliency map, the patch selection module does not select any patch on the highlighted regions.
A.S.
Thanks for that,
Struggling with dependencies. What python version are you using? are you using virtualenv? Are you able to remake a fresh environment? Depending on the python version, I'm having trouble with deprecations on Ubuntu 20.04.
This is the OS info that we primarily run experients on: Operating System: Red Hat Enterprise Linux CPE OS Name: cpe:/o:redhat:enterprise_linux:7.6:GA:server Kernel: Linux 4.14.0-115.33.1.el7a.ppc64le Architecture: ppc64-le
I think both Python 3.6.x and 3.7.x should work. I believe the main reason is PyTorch version. If you have an empty env, then pip install -r requirements.txt should do the work.
Thanks @seyiqi . There are some sym link issues with virtualenv on ubuntu 20.04 and using sudo
defaults to python3.8. Works with torch 1.1.0.
max_idx_x = max_idx_x.type(torch.long) # max_idx_x is float in torch 1.8
Needs to go in above line 199 in tools.get_max_window for torch 1.8
On Sun, 28 Mar 2021, 17:50 Yuanhong Chen, @.***> wrote:
I think the reason we always get x=0 is due to the following code in 'scr/utilities/tools.py', max_idx_x = max_linear_idx / W_map max_idx_y = max_linear_idx - max_idx_x * W_map max_idx_y here will always output 0
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/nyukat/GMIC/issues/13#issuecomment-808858787, or unsubscribe https://github.com/notifications/unsubscribe-auth/AIAOZR3JODTQM5V6AEFC4Q3TF3KCPANCNFSM4XVRFYQQ .
Hey,
see visualisation for 0_R-MLO.png![image](https://user-images.githubusercontent.com/33615047/108005245-1bb90a80-7048-11eb-89f1-fe60673a5463.png)
This occurs for only some ROIs eg 0_L-MLO.png:![image](https://user-images.githubusercontent.com/33615047/108004990-5e2e1780-7047-11eb-9a84-002d306d52de.png)
I'm exploring around here but any pointers would be much appreciated.
Also, the max value approach will be counfounded by markers, and probably perform poorly for women with very dense breasts eg:![image](https://user-images.githubusercontent.com/33615047/108005094-a5b4a380-7047-11eb-846e-eadd9fb8e4db.png)
In my case, all patches can't get past x=0 (stuck on the left border).
Is this just me?