Overview I'm trying to create a model that takes a "foot heatmap" (input image) and predicts a "shell heatmap" (true heatmap). My data contains foot heatmaps with a corresponding shell heatmap.
What I've done so far
- Basic Linear regression model using scikit-learn. The output is OK but I need it to be more precise.
- Neural network model (really off) - I'm not sure why the predicted image looks more like the input image ...
My question/ help
How can I improve my model? How do I get a proper image output using a CNN?
Any comment/discussion is appreciated - I haven't found any projects similar to this
Please note that I'm not looking to make a hybrid of the two images but instead I want my model to produce a shell image based on a foot input image (so no Transfering Learning or StyleGAN or pix2pix)