98 lines
2.8 KiB
Python
98 lines
2.8 KiB
Python
|
"""
|
||
|
|
author: ruben van de ven
|
||
|
|
"""
|
||
|
|
|
||
|
|
import ast
|
||
|
|
import os
|
||
|
|
import fire
|
||
|
|
from pathlib import Path
|
||
|
|
from detection.data import get_dataset
|
||
|
|
import pandas as pd
|
||
|
|
import PIL
|
||
|
|
from PIL import Image, ImageDraw
|
||
|
|
import logging
|
||
|
|
import coloredlogs
|
||
|
|
|
||
|
|
from detection.data.base import BaseDataset
|
||
|
|
from detection.data.info import DatasetInfo
|
||
|
|
|
||
|
|
coloredlogs.install()
|
||
|
|
logger = logging.getLogger(__name__)
|
||
|
|
|
||
|
|
# inline get_dataset("train") because of hard coded paths
|
||
|
|
|
||
|
|
|
||
|
|
split = "train"
|
||
|
|
meta = pd.read_csv("./data/meta.csv")
|
||
|
|
meta['image_path'] = [f'data/image/{panoid}_{heading}.jpg' for panoid, heading in zip(meta['panoid'], meta['heading'])]
|
||
|
|
|
||
|
|
info = DatasetInfo.load("./data/info.yaml")
|
||
|
|
duplicates = pd.read_csv(
|
||
|
|
"./data/duplicates.txt",
|
||
|
|
sep=' ',
|
||
|
|
names=['hash', 'path'])
|
||
|
|
# duplicates['panoid'] = duplicates['path'].str[11:-4]
|
||
|
|
|
||
|
|
meta = meta[~meta['image_path'].isin(duplicates['path'])]
|
||
|
|
|
||
|
|
dataset = BaseDataset(info, meta)[split]
|
||
|
|
|
||
|
|
|
||
|
|
def save_non_empty():
|
||
|
|
meta.to_csv("./data/non-empty-meta.csv")
|
||
|
|
print("Saved to", "./data/non-empty-meta.csv")
|
||
|
|
|
||
|
|
|
||
|
|
def render_images(dirname="/tmp/surveilling-surveillance/annotations"):
|
||
|
|
save_dir = Path(dirname)
|
||
|
|
save_dir.mkdir(parents=True, exist_ok=True)
|
||
|
|
|
||
|
|
#generally used with dataset.detection_dataloader(...)
|
||
|
|
print(meta)
|
||
|
|
# filter non empty on a stringified list
|
||
|
|
non_empty = meta[meta['annotations'] != "[]"]
|
||
|
|
|
||
|
|
|
||
|
|
stat_images = 0
|
||
|
|
stat_annotated_images = 0
|
||
|
|
stat_annotations = 0
|
||
|
|
|
||
|
|
for i, entry in non_empty.iterrows():
|
||
|
|
# category_id, bbox, bbox_mode, segmentation
|
||
|
|
annotations = ast.literal_eval(entry.annotations)
|
||
|
|
image_path = entry.image_path # f'./data/image/{entry.panoid}_{entry.heading}.jpg'
|
||
|
|
|
||
|
|
# print([a['bbox'] for a in annotations],os.path.exists(image_path))
|
||
|
|
try:
|
||
|
|
image = Image.open(image_path)
|
||
|
|
except PIL.UnidentifiedImageError:
|
||
|
|
logger.error(f"Invalid image: {image_path}")
|
||
|
|
continue
|
||
|
|
|
||
|
|
stat_images += 1
|
||
|
|
|
||
|
|
for y, ann in enumerate(annotations):
|
||
|
|
draw = ImageDraw.Draw(image)
|
||
|
|
draw.rectangle(ann['bbox'], outline='red', width=2)
|
||
|
|
crop = image.crop(ann['bbox'])
|
||
|
|
|
||
|
|
if 0 in crop.size:
|
||
|
|
logger.warning(f"Invalid crop {crop.size} using bbox {ann['bbox']} in {image_path}")
|
||
|
|
continue
|
||
|
|
|
||
|
|
stat_annotations += 1
|
||
|
|
if y == 0:
|
||
|
|
stat_annotated_images += 1
|
||
|
|
|
||
|
|
crop.save(save_dir / f'crop-{entry.panoid}_{entry.heading}_{y}.jpg')
|
||
|
|
|
||
|
|
image.save(save_dir / f'{entry.panoid}_{entry.heading}.jpg')
|
||
|
|
|
||
|
|
|
||
|
|
print(f"Total {stat_images} images and {stat_annotations} annotations.")
|
||
|
|
print(f"{stat_annotated_images} images with annotations, {stat_images - stat_annotated_images} images without annotations")
|
||
|
|
|
||
|
|
|
||
|
|
if __name__ == "__main__":
|
||
|
|
fire.Fire()
|