| import numpy as np, torch | |
| from PIL import Image | |
| import sys | |
| RAE_OLD="/mnt/tidal-alsh-share2/dataset/qinshengqian/research/c3/Code/RAE" | |
| sys.path.insert(0, RAE_OLD+"/src") | |
| from eval.fid import calculate_rfid | |
| ref = np.load(RAE_OLD+"/data/isic2019_imagefolder/val_224.npz")["arr_0"] | |
| print("ref", ref.shape, "std=%.0f"%ref.std(), flush=True) | |
| for a in ["dinov3l-k1","dinov3l-k23","sdvae","indomainvae"]: | |
| g = np.load(f"results/stage2/eval/gen_{a}.npz")["arr_0"] | |
| g224 = np.stack([np.asarray(Image.fromarray(x).resize((224,224),Image.BICUBIC)) for x in g]).astype(np.uint8) | |
| fid = calculate_rfid(ref, g224, 128, "cuda") | |
| print(f"FID {a}: {fid:.2f} (gen std=%.0f)"%g.std(), flush=True) | |
| print("FID_ALL_DONE", flush=True) | |