codewithdark

/

vit-chest-xray

like 4

Bro, you can find it here..

Kaggle notebook >> https://www.kaggle.com/code/codewithdark/vit-chest-xray
Github Code >> https://github.com/TheSmartTechnologyLab/vit-chest-xray

oh thank you a lot, i juat need the heatmaps (most interest area) per images so i needed the code, thank you a lot, also i have a question, you are saying it is 98% Accuracy or AUC when i test it on data that it does not exist in your dataset, the model makes false classifications I'm sorry to let you hear that, I'm surprised too.

When you test the model on data outside of the training distribution, it’s expected to produce incorrect classifications — the model simply hasn’t seen that class during training.

What I actually need from the implementation are per-image heatmaps that highlight the most relevant regions (areas the model is focusing on). That’s the main deliverable I’m looking for.

I haven’t integrated Grad-CAM yet for visual explanations. But if you’d like Grad-CAM overlays on the images as well, I can include that in the solution.

that exactly what i want, it is the areas that the model focuses on, which is the same goal you want, so did you got correct/high quality outputs from your model ?! in any way not just the Grad-CAM ! Can i message you in a social media ? like to talk more we may have same goals . if you want my telegram username is @iqcum . i want to hear what did you achieved from this model and what is your goals. I'm sorry for the weak English.

Here is my social Accounts > https://linktr.ee/codewithdark

HI, did you achieve the "heatmap per image"?