Source code, Models, and Datasets for our paper - AdaptDCT: A Deep Learning-Guided Multi-Scale DCT Compression Framework for Edge Devices
Clone the directory, run src/02-train to train the model, src/03-qat to quantise it, and tfliteconvert and tflite_infer to run it on the coral tpu. prune.py prunes the model and trimprune removes the sparse nodes.
Documentation and the paper will be added soon.