"""Evaluate an off-the-shelf DeiT-III (no KoPE) timm checkpoint on the same
materialized ImageNet-1K val ImageFolder tree, with the same eval protocol
(resize/center-crop/normalize) as simdinov2's eval.py, for an apples-to-apples
comparison against the released KoPE checkpoints.
"""
import argparse
import json
import time

import torch
import timm
from torch.utils.data import DataLoader
from torchvision import datasets, transforms
from torchvision.transforms import InterpolationMode

IMAGENET_DEFAULT_MEAN = (0.485, 0.456, 0.406)
IMAGENET_DEFAULT_STD = (0.229, 0.224, 0.225)


def build_transform(image_size=224, crop_pct=1.0):
    resize_size = int(round(image_size / crop_pct))
    return transforms.Compose([
        transforms.Resize(resize_size, interpolation=InterpolationMode.BICUBIC),
        transforms.CenterCrop(image_size),
        transforms.ToTensor(),
        transforms.Normalize(IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD),
    ])


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--model-name", required=True)
    ap.add_argument("--imagenet-val", required=True)
    ap.add_argument("--output-json", required=True)
    ap.add_argument("--batch-size", type=int, default=256)
    ap.add_argument("--crop-pct", type=float, default=1.0)
    args = ap.parse_args()

    device = "cuda" if torch.cuda.is_available() else "cpu"
    model = timm.create_model(args.model_name, pretrained=True).to(device).eval()

    transform = build_transform(224, args.crop_pct)
    dataset = datasets.ImageFolder(args.imagenet_val, transform=transform)
    loader = DataLoader(dataset, batch_size=args.batch_size, shuffle=False,
                         num_workers=8, pin_memory=True)

    top1, top5, n = 0, 0, 0
    t0 = time.time()
    with torch.no_grad():
        for imgs, targets in loader:
            imgs, targets = imgs.to(device, non_blocking=True), targets.to(device, non_blocking=True)
            with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=(device == "cuda")):
                logits = model(imgs)
            _, pred5 = logits.topk(5, dim=1)
            correct = pred5.eq(targets.unsqueeze(1))
            top1 += correct[:, 0].sum().item()
            top5 += correct.any(dim=1).sum().item()
            n += targets.size(0)

    result = {
        "model_name": args.model_name,
        "n_images": n,
        "top1_acc": 100.0 * top1 / n,
        "top5_acc": 100.0 * top5 / n,
        "eval_seconds": time.time() - t0,
    }
    print(json.dumps(result, indent=2))
    with open(args.output_json, "w") as f:
        json.dump(result, f, indent=2)


if __name__ == "__main__":
    main()