Q-Study HSCI

This project trains hyperspectral image classification models, evaluates the saved checkpoint, and optionally quantizes the model with HQQ or Quanto.

Quick Start

1. Train a model first.

python train.py --model mvit --dataset UP

2. Evaluate the saved model and quantize it.

python eval.py --model mvit --dataset UP --quant_method hqq --nbits 8 --group_size 64

If you want to compare the original and quantized model with CKA, enable the --cka flag:

python eval.py --model mvit --dataset UP --quant_method hqq --nbits 8 --group_size 64 --cka True

Notes

• train.py contains the training arguments.
• eval.py contains evaluation, quantization, and comparison arguments.
• For Indian Pines, the training split is capped at 10 samples per class in the code.