code/scripts/generate_all.py

"""
Batch generation for all medical datasets.
For each dataset, generates images conditioned on ALL masks, saves individually.
Records per-image sampling time.

Usage:
  python scripts/generate_all.py --dataset cvc
  python scripts/generate_all.py --dataset kvasir
  python scripts/generate_all.py --dataset refuge2
  python scripts/generate_all.py --dataset all
"""
import sys
sys.path.insert(0, "/data/sichengli/Code/PixelGen")

import argparse
import os
import gc
import time
import json
import random
import numpy as np
import torch
from PIL import Image
import torchvision.transforms as transforms
import torchvision.transforms.functional as TF
from torch.utils.data import Dataset, DataLoader

from src.models.transformer.JiT_medical import JiTMedical


# ─── Config ───────────────────────────────────────────────────────────
CONFIGS = {
    "cvc": {
        "data_root": "/data2/sichengli/Data/test/Segmentation/CVC-ClinicDB",
        "img_subdir": "PNG/Original",
        "mask_subdir": "PNG/Ground Truth",
        "file_ext": (".png",),
        "ckpt": "/data/sichengli/Code/PixelGen/medical_workdirs/exp_PixelGen_Medical_CVC/epoch=19999-step=100000.ckpt",
        "out_dir": "/data/sichengli/Code/PixelGen/medical_workdirs/exp_PixelGen_Medical_CVC/generated_images",
        "multi_split": False,
        "train_ratio": 0.9,
        "seed": 42,
    },
    "kvasir": {
        "data_root": "/data2/sichengli/Data/test/Segmentation/Kvasir-SEG/Kvasir-SEG",
        "img_subdir": "images",
        "mask_subdir": "masks",
        "file_ext": (".jpg", ".png", ".jpeg"),
        "ckpt": "/data/sichengli/Code/PixelGen/medical_workdirs/exp_PixelGen_Medical_Kvasir/epoch=12499-step=100000.ckpt",
        "out_dir": "/data/sichengli/Code/PixelGen/medical_workdirs/exp_PixelGen_Medical_Kvasir/generated_images",
        "multi_split": False,
        "train_ratio": 0.9,
        "seed": 42,
    },
    "refuge2": {
        "data_root": "/data2/sichengli/Data/test/Segmentation/REFUGE2",
        "splits": ["train", "val", "test"],
        "file_ext": (".jpg", ".png", ".jpeg"),
        "mask_ext": (".bmp", ".png"),
        "ckpt": "/data/sichengli/Code/PixelGen/medical_workdirs/exp_PixelGen_Medical_REFUGE2/epoch=16666-step=100000.ckpt",
        "out_dir": "/data/sichengli/Code/PixelGen/medical_workdirs/exp_PixelGen_Medical_REFUGE2/generated_images",
        "multi_split": True,
        "val_ratio": 0.1,
        "seed": 42,
    },
}

MODEL_KWARGS = dict(
    input_size=256, patch_size=16, in_channels=3,
    hidden_size=768, depth=12, num_heads=12, mlp_ratio=4.0,
    attn_drop=0.0, proj_drop=0.1, num_classes=1,
    use_bottleneck=True, bottleneck_dim=128,
    in_context_len=32, in_context_start=4, mask_in_channels=1,
    mask_mode="spatial"
)

RESOLUTION = 256
BATCH_SIZE = 16
NUM_STEPS = 50
CFG_SCALE = 2.0


# ─── Dataset ──────────────────────────────────────────────────────────
class MaskDataset(Dataset):
    """Load all masks from a dataset for generation."""
    def __init__(self, cfg):
        self.resolution = RESOLUTION
        self.pairs = []  # (mask_path, save_name)

        if cfg.get("multi_split"):
            for split in cfg["splits"]:
                img_dir = os.path.join(cfg["data_root"], split, "images")
                mask_dir = os.path.join(cfg["data_root"], split, "mask")
                img_files = sorted([f for f in os.listdir(img_dir) if f.endswith(cfg["file_ext"])])
                for img_f in img_files:
                    base_name = os.path.splitext(img_f)[0]
                    for ext in cfg["mask_ext"]:
                        candidate = os.path.join(mask_dir, base_name + ext)
                        if os.path.exists(candidate):
                            save_name = f"{split}_{base_name}"
                            self.pairs.append((candidate, save_name))
                            break
        else:
            mask_dir = os.path.join(cfg["data_root"], cfg["mask_subdir"])
            all_files = sorted([f for f in os.listdir(mask_dir) if f.endswith(cfg["file_ext"])])
            for f in all_files:
                base_name = os.path.splitext(f)[0]
                self.pairs.append((os.path.join(mask_dir, f), base_name))

        print(f"[MaskDataset] {len(self.pairs)} masks loaded")

    def __len__(self):
        return len(self.pairs)

    def __getitem__(self, idx):
        mask_path, save_name = self.pairs[idx]
        mask = Image.open(mask_path).convert("L")
        mask = TF.resize(mask, (self.resolution, self.resolution),
                        interpolation=transforms.InterpolationMode.NEAREST)
        mask_tensor = TF.to_tensor(mask)
        return mask_tensor, save_name


# ─── Sampling ─────────────────────────────────────────────────────────
def shift_respace_fn(t, shift=1.0):
    return t / (t + (1 - t) * shift)


@torch.no_grad()
def sample_batch_cfg(model, noise, mask, num_steps=50, cfg_scale=2.0, t_eps=0.05):
    """
    Euler ODE sampler with Classifier-Free Guidance.
    - Scheduler: Linear (t goes from 0 to 1)
    - Timeshift: 1.0 (no shift)
    - Prediction: x0-prediction, converted to velocity v = (x0 - x_t) / (1-t)
    - CFG: v = v_uncond + cfg_scale * (v_cond - v_uncond)
    """
    batch_size = noise.shape[0]
    timesteps = torch.linspace(0.0, 1 - 1.0 / num_steps, num_steps)
    timesteps = torch.cat([timesteps, torch.tensor([1.0])], dim=0)
    timesteps = shift_respace_fn(timesteps, 1.0).to(noise.device)
    y = torch.zeros(batch_size, dtype=torch.long, device=noise.device)
    x = noise
    for i in range(len(timesteps) - 1):
        t_cur = timesteps[i]
        t_next = timesteps[i + 1]
        dt = t_next - t_cur
        t_batch = t_cur.repeat(batch_size)
        # CFG: concat uncond (mask=0) and cond (mask=real)
        cfg_x = torch.cat([x, x], dim=0)
        cfg_t = t_batch.repeat(2)
        cfg_y = torch.cat([y, y], dim=0)
        cfg_mask = torch.cat([torch.zeros_like(mask), mask], dim=0)
        pred = model(cfg_x, cfg_t, cfg_y, mask=cfg_mask)
        pred_v = (pred - cfg_x) / (1.0 - cfg_t.view(-1, 1, 1, 1)).clamp_min(t_eps)
        v_uncond, v_cond = pred_v.chunk(2)
        v = v_uncond + cfg_scale * (v_cond - v_uncond)
        x = x + v * dt
    return x


def load_model(ckpt_path, device):
    ckpt = torch.load(ckpt_path, map_location="cpu", weights_only=False)
    state_dict = ckpt["state_dict"]
    ema_state = {}
    for k, v in state_dict.items():
        if k.startswith("ema_denoiser."):
            new_k = k.replace("ema_denoiser.", "").replace("_orig_mod.", "")
            ema_state[new_k] = v
    model = JiTMedical(**MODEL_KWARGS)
    result = model.load_state_dict(ema_state, strict=False)
    print(f"Loaded EMA ({len(ema_state)} keys), missing: {result.missing_keys}, unexpected: {result.unexpected_keys}")
    model = model.to(device).eval().to(torch.float32)
    return model


# ─── Main ─────────────────────────────────────────────────────────────
def generate(dataset_name):
    cfg = CONFIGS[dataset_name]
    device = torch.device("cuda:0")

    out_dir = cfg["out_dir"]
    os.makedirs(out_dir, exist_ok=True)

    print(f"\n{'='*60}")
    print(f"  Generating: {dataset_name.upper()}")
    print(f"  Sampling: Euler ODE + CFG={CFG_SCALE}, {NUM_STEPS} steps")
    print(f"  Output: {out_dir}")
    print(f"{'='*60}\n")

    # Load dataset
    dataset = MaskDataset(cfg)
    loader = DataLoader(dataset, batch_size=BATCH_SIZE, shuffle=False,
                       num_workers=4, pin_memory=True)

    # Load model
    print("Loading model...")
    model = load_model(cfg["ckpt"], device)

    # Generate
    torch.manual_seed(0)
    timing_records = []
    total_images = 0
    total_time = 0.0

    for batch_idx, (masks, save_names) in enumerate(loader):
        bs = masks.shape[0]
        masks = masks.to(device)
        noise = torch.randn(bs, 3, RESOLUTION, RESOLUTION, device=device)

        # Time the sampling
        torch.cuda.synchronize()
        t_start = time.time()

        gen = sample_batch_cfg(model, noise, masks, NUM_STEPS, CFG_SCALE)

        torch.cuda.synchronize()
        t_end = time.time()

        batch_time = t_end - t_start
        per_image_time = batch_time / bs
        total_time += batch_time
        total_images += bs

        # Clamp and convert to [0, 255]
        gen = gen.clamp(-1, 1) * 0.5 + 0.5  # [-1,1] -> [0,1]

        # Save each image
        for i in range(bs):
            img_np = (gen[i].permute(1, 2, 0).cpu().numpy() * 255).clip(0, 255).astype(np.uint8)
            save_path = os.path.join(out_dir, f"{save_names[i]}.png")
            Image.fromarray(img_np).save(save_path)

            timing_records.append({
                "filename": f"{save_names[i]}.png",
                "time_seconds": per_image_time,
            })

        print(f"  Batch {batch_idx+1}/{len(loader)} | {bs} images | {batch_time:.2f}s ({per_image_time:.3f}s/img)")

    avg_time = total_time / total_images
    print(f"\nGeneration complete: {total_images} images")
    print(f"Total time: {total_time:.2f}s")
    print(f"Average per-image: {avg_time:.4f}s ({1.0/avg_time:.1f} img/s)")

    # Save timing summary
    summary = {
        "dataset": dataset_name,
        "num_images": total_images,
        "sampling_strategy": "Euler ODE (1st-order)",
        "num_steps": NUM_STEPS,
        "cfg_scale": CFG_SCALE,
        "scheduler": "LinearScheduler",
        "timeshift": 1.0,
        "resolution": RESOLUTION,
        "total_time_seconds": round(total_time, 4),
        "avg_time_per_image_seconds": round(avg_time, 4),
        "throughput_img_per_sec": round(1.0 / avg_time, 2),
        "per_image_timing": timing_records,
    }

    summary_path = os.path.join(out_dir, "generation_stats.json")
    with open(summary_path, "w") as f:
        json.dump(summary, f, indent=2)
    print(f"Stats saved: {summary_path}")

    return total_images, avg_time


if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--dataset", type=str, required=True,
                       choices=["cvc", "kvasir", "refuge2", "all"])
    args = parser.parse_args()

    datasets = ["cvc", "kvasir", "refuge2"] if args.dataset == "all" else [args.dataset]

    all_results = {}
    for ds in datasets:
        n_imgs, avg_t = generate(ds)
        all_results[ds] = (n_imgs, avg_t)
        gc.collect()
        torch.cuda.empty_cache()

    # Final summary
    print(f"\n{'='*60}")
    print(f"  GENERATION SUMMARY")
    print(f"  Sampling: Euler ODE, {NUM_STEPS} steps, CFG={CFG_SCALE}")
    print(f"{'='*60}")
    print(f"{'Dataset':<15s} | {'Images':>8s} | {'Avg Time':>10s} | {'Throughput':>12s}")
    print("-" * 55)
    for name, (n, t) in all_results.items():
        print(f"{name:<15s} | {n:>8d} | {t:>8.4f}s | {1.0/t:>8.1f} img/s")
    print("=" * 55)