import os
from os import path
import glob
import numpy as np
from PIL import Image
from diffusers.utils import export_to_gif

import torch

ROOT = '/projects/katefgroup/datasets/gs_kubric/InpaintingFormat_Set50'
OUT_ROOT = os.path.join(ROOT, 'Consistent4D')
RESOLUTION = 512
imset = '2017/consistent4d.txt'

os.makedirs(OUT_ROOT, exist_ok=True)

device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")

mask_dir = path.join(ROOT, 'MergedAnnotations')
image_dir = path.join(ROOT, 'JPEGImages')

videos = []
num_frames = {}

vid_names = os.listdir(image_dir)
for _video in vid_names:
    videos.append(_video)
    num_frames[_video] = len(os.listdir(path.join(image_dir, _video)))

for index in range(len(videos)):

    video = videos[index]
    if video == 'swing-human-only':
        continue

    all_frames = []

    img_paths = glob.glob(path.join(image_dir, video + '/*'))
    img_paths.sort()

    mask_paths = glob.glob(path.join(mask_dir, video, '001' + '/*'))
    mask_paths.sort()

    os.makedirs(os.path.join(OUT_ROOT, video), exist_ok=True)

    for f in range(10):
        
        # Load RGBs
        img = Image.open(img_paths[f]).convert('RGBA')
        img = img.resize((RESOLUTION, RESOLUTION))

        mask = Image.open(mask_paths[f])
        mask = mask.resize((RESOLUTION, RESOLUTION))
        binary_mask = np.array(mask) > 128
        binary_mask = np.tile(binary_mask[:, :, np.newaxis], (1,1,4))
        
        masked_img = Image.fromarray(np.array(img) * binary_mask).convert('RGBA')
        masked_img.save(os.path.join(OUT_ROOT, video, f'{f}.png'))