fish-vista / download_and_process_nd_images.py
ksmehrab's picture
PR to add ND processing files and script (#3)
7804064 verified
raw
history blame
2.4 kB
import pandas as pd
import requests
from PIL import Image
from io import BytesIO
import os
from tqdm import tqdm
import json
import argparse
parser = argparse.ArgumentParser(description='Download, process and save ND images')
parser.add_argument('--save_dir', type=str, help='The directory where processed images will be saved')
args = parser.parse_args()
save_dir = args.save_dir
if not os.path.exists('ND_Processing_Files'):
raise Exception('The ND processing file directory is not found in your current working directory')
if not os.path.exists(save_dir):
os.mkdir(save_dir)
nd_df = pd.read_csv(os.path.join('ND_Processing_Files', 'ND_data.csv'))
with open(os.path.join('ND_Processing_Files', 'nd_filenames_bboxes_map.json'), 'r') as f:
nd_filenames_bboxes_map = json.load(f)
seg_mask_dir = os.path.join('ND_Processing_Files', 'ND_background_masks')
padding = 20
for i, row in tqdm(nd_df.iterrows()):
target_filename = row['filename']
download_url = row['original_url']
local_filename = 'temp.jpg'
response = requests.get(download_url)
# Ensure the request was successful
response.raise_for_status()
# Convert response content to a PIL Image
image = Image.open(BytesIO(response.content))
target_bbox = nd_filenames_bboxes_map[target_filename]
# crop the image
left, upper, right, lower = target_bbox
max_width, max_height = image.size
padded_left = max(left - padding, 0)
padded_upper = max(upper - padding, 0)
padded_right = min(right + padding, max_width)
padded_lower = min(lower + padding, max_height)
# Crop the image using the adjusted, padded bounding box
cropped_image = image.crop((padded_left, padded_upper, padded_right, padded_lower))
assert target_filename[-4] == '.', 'The code assumes we have .<extension> at the end'
target_seg_mask_file = target_filename[:-4]+'.png'
if target_seg_mask_file in os.listdir(seg_mask_dir):
mask_image = Image.open(os.path.join(seg_mask_dir, target_seg_mask_file)).convert('L')
else:
print(f'Segmentation mask not found for target image {target_filename}. Skipping...')
continue
white_image = Image.new("RGB", cropped_image.size, (255, 255, 255))
result_image = Image.composite(cropped_image, white_image, mask_image)
result_image.save(os.path.join(save_dir, target_filename))