There are some queries regarding image based competition on kaggle. Let's say after creating a baseline model for objectDetection/imageClassification/poseEstimation or any other problem. Now the point is how to improve that model.
- Do I need an understanding of Image processing or should I focus on model hyper parameter.
- If image processing is required then should I go for image understanding using Fourier transform and all. And if I do then how can I apply these concept in high level kernels.
- If I go for some processing library then can anyone provide some resources where I can get some understanding regarding that.
- And If I focus on model hyper parameter tuning than any advice is preferred as it will take a lot of time just to check if some parameter are working or not.
I would be grateful for any type of advice or tips that you can provide.