#!/bin/bash

# OCTA500 训练 (已完成, mask_mode=global, 最佳权重: epoch=236-step=100000.ckpt)
# cd /data/sichengli/Code/PixelGen && WANDB_MODE=offline python3 main.py fit --config configs_medical/PixelGen_Medical_B16.yaml 2>&1 | tee training_medical_b16.log

# CVC-ClinicDB 训练 (已完成, mask_mode=spatial, 最佳权重: epoch=19999-step=100000.ckpt)
# cd /data/sichengli/Code/PixelGen && torchrun --nproc_per_node=8 main.py fit --config configs_medical/PixelGen_Medical_CVC.yaml 2>&1 | tee training_medical_cvc.log

# Kvasir-SEG 训练 (已完成, mask_mode=spatial, 最佳权重: epoch=12499-step=100000.ckpt)
# cd /data/sichengli/Code/PixelGen && torchrun --nproc_per_node=8 main.py fit --config configs_medical/PixelGen_Medical_Kvasir.yaml 2>&1 | tee training_medical_kvasir.log

# REFUGE2 训练 (mask_mode=spatial, 100k steps, 3-class fundus masks)
cd /data/sichengli/Code/PixelGen && torchrun --nproc_per_node=8 main.py fit --config configs_medical/PixelGen_Medical_REFUGE2.yaml 2>&1 | tee training_medical_refuge2.log

# CVC-ClinicDB 评估 (FID/Precision/Recall)
# cd /data/sichengli/Code/PixelGen && CUDA_VISIBLE_DEVICES=0 python scripts/evaluate_medical.py --dataset cvc --cfg

# Kvasir-SEG 评估 (FID/Precision/Recall)
# cd /data/sichengli/Code/PixelGen && CUDA_VISIBLE_DEVICES=0 python scripts/evaluate_medical.py --dataset kvasir --cfg

# REFUGE2 评估 (FID/Precision/Recall)
# cd /data/sichengli/Code/PixelGen && CUDA_VISIBLE_DEVICES=0 python scripts/evaluate_medical.py --dataset refuge2 --cfg