kvasir-points_datasets_script.py

# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""Dataset for filtered Kvasir-instrument and Hyper-Kvasir with bounding boxes."""

import os
import io
import json
from PIL import Image
import datasets

import os
import json
import pandas as pd
import hashlib  
from collections import defaultdict
import numpy as np



# def cal_mid_xy(bx): return [{"x": float(box['xmin'] + box['xmax']) / 2,
#                              "y": float(box['ymin'] + box['ymax']) / 2} for box in bx]


def cal_sha256(file_path): return hashlib.sha256(
    open(file_path, 'rb').read()).hexdigest()


def convert_to_json_format(file_path, image_width, image_height):
    with open(file_path, 'r') as file:
        return [
            {
                "label": line.split()[0],
                "xmin": int((float(line.split()[1]) - float(line.split()[3]) / 2) * image_width),
                "ymin": int((float(line.split()[2]) - float(line.split()[4]) / 2) * image_height),
                "xmax": int((float(line.split()[1]) + float(line.split()[3]) / 2) * image_width),
                "ymax": int((float(line.split()[2]) + float(line.split()[4]) / 2) * image_height),
            }
            for line in file.readlines()
        ]
    
def get_image_bytes(img_path, max_width=700):
    img = Image.open(img_path)
    if img.width <= max_width:
        return open(img_path, "rb").read(), img.width, img.height, 1.0
    with io.BytesIO() as b:
        img.resize((max_width, int(img.height * max_width / img.width))).save(b, "PNG")
        return b.getvalue(), max_width, int(img.height * max_width / img.width), float(max_width) / img.width


def get_bboxes(bx, ratio):
    return [[box[k] * ratio for k in ('xmin', 'ymin', 'xmax', 'ymax')] for box in bx]


def cal_mid(bx, ratio): 
    return [[((box['xmin'] + box['xmax']) / 2) * ratio,
             ((box['ymin'] + box['ymax']) / 2) * ratio] for box in bx]


class_map = {"0": "normal",  "1": "cluster", "2": "pinhead"}

hyper_label_img_path = '/global/D1/projects/HOST/Datasets/hyper-kvasir/labeled-images/image-labels.csv'

hyper_df = pd.read_csv(hyper_label_img_path)

hyper_seg_img_path = '/global/D1/projects/HOST/Datasets/hyper-kvasir/segmented-images/bounding-boxes.json'
hyper_seg_img_base_path = "/global/D1/projects/HOST/Datasets/hyper-kvasir/segmented-images/images"

instr_seg_img_path = '/global/D1/projects/HOST/Datasets/kvasir-instrument/bboxes.json'
instr_seg_img_base_path = '/global/D1/projects/HOST/Datasets/kvasir-instrument/images/'

hyper_seg_imgs = json.load(open(hyper_seg_img_path))
instr_seg_imgs = json.load(open(instr_seg_img_path))

visem_root = "/global/D1/projects/HOST/Datasets/visem-tracking"

_CITATION = """\
@article{kvasir,
    title={Kvasir-instrument and Hyper-Kvasir datasets for bounding box annotations},
    author={Sushant Gautam and collaborators},
    year={2024}
}
"""

_DESCRIPTION = """
Filtered Kvasir-instrument and Hyper-Kvasir datasets with bounding boxes for medical imaging tasks. 
Each entry contains images, bounding box coordinates, and additional metadata.
"""

_HOMEPAGE = "https://example.com/kvasir-hyper-bbox"

_LICENSE = "CC BY-NC 4.0"

_URLS = {
    "filtered_data": "https://example.com/kvasir-hyper-bbox-dataset.zip"
}


class KvasirHyperBBox(datasets.GeneratorBasedBuilder):
    """Dataset for Kvasir-instrument and Hyper-Kvasir with bounding boxes."""

    VERSION = datasets.Version("1.0.0")

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name="bbox_dataset",
            version=VERSION,
            description="Dataset with bounding box annotations."
        )
    ]

    DEFAULT_CONFIG_NAME = "bbox_dataset"

    def _info(self):
        features = datasets.Features({
            "image_data": datasets.Image(),
            "image_sha256": datasets.Value("string"),
            "img_size": datasets.Sequence(datasets.Value("float32")),
            "points": datasets.Sequence(datasets.Sequence(datasets.Value("float32"))),
            "bbox": datasets.Sequence(datasets.Sequence(datasets.Value("float32"))),
            "count": datasets.Value("int64"),
            "label": datasets.Value("string"),
            "collection_method": datasets.Value("string"),
            "classification": datasets.Value("string"),
            "organ": datasets.Value("string")
        })

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
            features=features
        )

    def _split_generators(self, dl_manager):
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={},
            )
        ]

    def _generate_examples(self):
        for key, entry in hyper_seg_imgs.items():
            img_path = os.path.join(hyper_seg_img_base_path, f"{key}.jpg")
            img, width, height, ratio = get_image_bytes(img_path)
            hyper_entry = hyper_df.loc[hyper_df['Video file'] == key].iloc[0]
            yield key, {
                "image_data": img,
                "image_sha256": cal_sha256(img_path),
                "img_size": [width, height],
                "points":  cal_mid(entry['bbox'], ratio),
                "bbox": get_bboxes(entry['bbox'], ratio),
                "count": len(entry['bbox']),
                "label": hyper_entry.Finding,
                "collection_method": 'counting',
                "classification": hyper_entry.Classification,
                "organ": hyper_entry.Organ
            }

        for key, entry in instr_seg_imgs.items():
            img_path = os.path.join(instr_seg_img_base_path, f"{key}.jpg")
            img, width, height, ratio = get_image_bytes(img_path)
            yield key, {
                "image_data": img,
                "image_sha256": cal_sha256(img_path),
                "img_size": [width, height],
                "points":  cal_mid(entry['bbox'], ratio),
                "bbox": get_bboxes(entry['bbox'], ratio),
                "count": len(entry['bbox']),
                "label": "instrument",
                "collection_method": "counting",
                "classification": "instrument",
                "organ": "instrument"
            }

        for folder in os.listdir(visem_root):
            folder_path = os.path.join(visem_root, folder)
            labels_all = os.listdir(folder_path+"/labels")
            labels = [labels_all[i] for i in np.linspace(
                0, len(labels_all)-1, 250).astype(int)]
            for label in labels:
                label_path = os.path.join(folder_path, "labels", label)
                image_path = label_path.replace(
                    "/labels/", "/images/").replace(".txt", ".jpg")
                img, width, height, ratio = get_image_bytes(image_path)
                entry_bbox = convert_to_json_format(label_path, width, height)
                label_dict = defaultdict(list)
                for entry in entry_bbox:
                    label_dict[entry['label']].append(entry)
                for label in label_dict:
                    yield cal_sha256(image_path)+label, {
                        "image_data": img,
                        "image_sha256": cal_sha256(image_path),
                        "img_size": [width, height],
                        "points": cal_mid(label_dict[label], ratio),
                        "bbox": get_bboxes(label_dict[label], ratio),
                        "count": len(label_dict[label]),
                        "label": class_map[label],
                        "collection_method": "counting",
                        "classification": "sperm",
                        "organ": "visem dataset"
                    }


# rm -rf /home/sushant/.cache/huggingface/modules/datasets_modules/datasets/kvasir-points_datasets_script/ /home/sushant/.cache/huggingface/datasets/kvasir-points_datasets_script 
# datasets-cli test  /global/D1/projects/HOST/Datasets/hyper-kvasir/sushant-experiments/kvasir-points_datasets_script.py --save_info  --all_configs --trust_remote_code
# huggingface-cli upload kvasir-points . . --repo-type dataset