Wenjun Hou (Ethan)

Hi @Markin-Wang, Thank you for your attention. I have uploaded the graph construction script for the MIMIC-CXR dataset which is similar to the one for IU X-ray, and you may...

Hi @Markin-Wang, Scripts of training on the MIMIC-CXR dataset have been uploaded and please check it out. Feel free to reopen this issue if you have any other relevant questions....

Hi, I have updated the code for extracting observation plans, and you can find it [here](https://github.com/wjhou/ORGan/blob/main/src/plan_extraction.py?plain=1#L197l) and you may add in your code or pull again. Basically, I just removed...

Hi, It sounds quite abnormal. 🤣 We conduct the experiments on an RTX 3090 GPU and Stage 1 takes about 1.5 hours to complete. RXT A6000 should have better performance...

Seems like the model can generate something and it is the pycoco issue. Can you check into details and see the outputs of the model for debugging?

We use the same annotation as you mentioned. It’s probably caused by the cider script. I remember I did not turn on the cider calculation.

Hmmm I am currently conducting some follow up experiments. Let me try some runs and I’ll get back to you later.

Well, it seems the model running properly and the issue might stem from the data preparation.

Hi, After reading my messy code again, I figured out why it takes so long to finish training. The images are loaded inside the collator instead of the dataset in...

Hi @Markin-Wang I have updated the code for better efficiency, i.e., moving the image loading part into the dataset. I test on my server and it works fine and you...