Function validate

train_s2.py:81–126

(
    ref_img_path,
    guid_folder,
    guid_types,
    guid_start_idx,
    clip_length,
    width, height,
    pipe,
    generator,
    denoising_steps=20,
    guidance_scale=3.5,
    aug_type="Resize",
)

def validate(
    ref_img_path,
    guid_folder,
    guid_types,
    guid_start_idx,
    clip_length,
    width, height,
    pipe,
    generator,
    denoising_steps=20,
    guidance_scale=3.5,
    aug_type="Resize",
):
    ref_img_pil = Image.open(ref_img_path)
    if aug_type =="Padding":
        ref_img_pil = padding_pil(ref_img_pil, height)
    guid_folder = Path(guid_folder)
    guid_img_pil_lst = []
    for guid_type in guid_types:
        guid_img_lst = sorted((guid_folder / guid_type).iterdir())
        guid_img_clip_lst = guid_img_lst[guid_start_idx: guid_start_idx + clip_length]
        single_guid_pil_lst = []
        for guid_img_path in guid_img_clip_lst:
            if guid_type == "semantic_map":
                mask_img_path = guid_folder / "mask" / guid_img_path.name
                guid_img_pil = mask_to_bkgd(guid_img_path, mask_img_path)
            else:
                guid_img_pil = Image.open(guid_img_path).convert("RGB")
            if aug_type == "Padding":
                guid_img_pil = padding_pil(guid_img_pil, height)
            single_guid_pil_lst += [guid_img_pil]
        guid_img_pil_lst.append(single_guid_pil_lst)
        
    val_videos = pipe(
        ref_img_pil,
        guid_img_pil_lst,
        guid_types,
        width,
        height,
        clip_length,
        denoising_steps,
        guidance_scale,
        generator=generator,
    ).videos
    
    return val_videos, ref_img_pil, guid_img_pil_lst


def log_validation(