I'm currently trying to train the smallest possible model for my object detection problem, based on yolov11n. I was wondering what is considered the best order to perform pruning, quantization and distillation.
My approach: I was thinking that I first need to train the base yolo model on my data, then perform pruning for each layer. Then distill this model (but with what base student model - I don't know). And finally export it with either FP16 or INT8 quantization, to ONNX or TFLite format.
Is this a good approach to minimize size/memory footprint while preserving performance? What would you do differently? Thanks for your help!
