#!/usr/bin/env python3

from pycocotools.coco import COCO
import os
import shutil
import json


# MODIFY ME!
# ---------------------------------
coco_dataset='/opt/datasets/coco/'
year='2017'
splits=['train', 'val']
dst_dir='data'
metadata='metadata.jsonl'
# ---------------------------------

def process_split(coco_dataset, year, split, dst_dir, metadata):
    annotations=f'{coco_dataset}/annotations/instances_{split}{year}.json'
    coco=COCO(annotations)
    cats =coco.loadCats(coco.getCatIds())
    
    target_name='cell phone'
    target_id=coco.getCatIds(catNms=[target_name])[0];
    
    print(f'Found "{target_name}" id: {target_id}')
    
    image_ids=coco.getImgIds(catIds=[target_id]);
    images=coco.loadImgs(image_ids)
    
    print(f'Found {len(images)} images containing a "{target_name}"')
    
    image_dir=f'{coco_dataset}/{split}{year}/'
    image_dst_dir=f'{dst_dir}/{split}/'
    
    os.makedirs(image_dst_dir, exist_ok=True)
    
    dst_metadata=f'{image_dst_dir}/{metadata}'
    with open(dst_metadata, 'w') as f:
        for image in images:
            ann_ids = coco.getAnnIds(imgIds=image['id'], catIds=target_id, iscrowd=None)
            anns = coco.loadAnns(ann_ids)
    
            src_image=f'{image_dir}/{image["file_name"]}'
            shutil.copy(src_image, image_dst_dir)
    
            entry={}
            entry['file_name']=image["file_name"]
            entry['image_id']=image["id"]
            boxes=[ann['bbox'] for ann in anns]
            cats=[0]*len(boxes)
            ids=[ann['id'] for ann in anns]
            areas=[ann['area'] for ann in anns]
            entry['objects']={
                'bbox': boxes,
                'category': cats,
                'id': ids,
                'area': areas
            }
    
            f.write(f'{json.dumps(entry)}\n')
    
    print(f'Done processing the "{split}" split!')

if __name__ == "__main__":
    for split in splits:
        process_split(coco_dataset, year, split, dst_dir, metadata)