Hi Asaf,
As per my understanding, CNN models are better suited for tasks like image classification and segmentation using dataset containing images. However, if you want to apply it to your handcrafted dataset follow the strategies given below:
- Determine if the dataset can be transformed into a two-dimensional format, such as a matrix or heatmap, and then apply convolutional layers to it.
- Try modifying the pretrained CNN to accept non-image data. CNN models generally expect image data, specifically 2D matrices with three color channels. But since the features are typically 1D vectors, change the first layer of model to accept a 1D vector instead of a 3D image. You might also need to replace some convolutional layers with dense (fully connected) layers that are better suited for 1D data. Then adjust the final layers to output the desired number of classes.
I hope this is beneficial.
