import json import datasets import pandas as pd id_to_original = { "1": "5-5-10-H-A1000C 100h-30k-3-crop", "2": "5-5-A1000C 100h-30k-9 crop", "3": "5-5-A1000C 100h-30k-9 crop2", "4": "5-5-A1000C 100h-30k-9-crop", "5": "5k-Cr-10-10-20Fe-H-Ageing1200C 4h-6-crop", "6": "Cr-5-5-10Fe-A1200C 4h-6 crop1", "7": "Cr-5-5-10Fe-A1200C 4h-6 crop2", "8": "Cr-5-5-10Fe-H1400-20h-A800-240h-80k-9crop1", "9": "Cr-5-5-10Fe-H1400-20h-A800-240h-80k-9crop2", "10": "Cr-5-5-10Fe-H1400-20h-A800-240h-80k-10 crop", "11": "Cr-5-5-10Fe-H1400-20h-A800-240h-80k-10 crop2", "12": "Cr-5-5-10Fe-H1400-20h-A1000-20h-50k-10 crop", "13": "Cr-5-5-10Fe-H1400-20h-A1000-240h-30k-8 crop2", "14": "Cr-5-5-A1200C 4h-20k-5-crop1", "15": "Cr-5-5-A1200C 4h-20k-5-crop2", "16": "Cr-10-10-20Fe-H20h-A1200C 20h-7-crop1", "17": "J955-H2-7-crop1", "18": "J955-H2-7-crop2", "19": "Cr-10-10-20Fe-A100h-1-crop1", "20": "Cr-10-10-20Fe-A100h-4-crop1", "21": "Cr-10Ni-10Al-20Fe-8 crop1", "22": "Cr-10Ni-10Al-20Fe-8 crop2", "23": "Cr-10Ni-10Al-20Fe-H1400C20h-9 crop1", "24": "Cr-10Ni-10Al-20Fe-H1400C20h-9 crop2", } ids_split = { datasets.Split.TEST: [ "1", "5", "9", "14", "20", ], datasets.Split.VALIDATION: [ "2", "7", "18", "22", ], datasets.Split.TRAIN: [ "3", "4", "6", "8", "10", "11", "12", "13", "15", "16", "17", "19", "21", "23", "24", ] } _CITATION = """\ @article{xia2023Accurate, author = {Zeyu Xia and Kan Ma and Sibo Cheng and Thomas Blackburn and Ziling Peng and Kewei Zhu and Weihang Zhang and Dunhui Xiao and Alexander J Knowles and Rossella Arcucci}, copyright = {CC BY-NC 3.0}, doi = {10.1039/d3cp00402c}, issn = {1463-9076}, journal = {Physical Chemistry Chemical Physics}, keywords = {}, language = {English}, month = {6}, number = {23}, pages = {15970--15987}, pmid = {37265373}, publisher = {Royal Society of Chemistry (RSC)}, title = {Accurate Identification and Measurement of the Precipitate Area by Two-Stage Deep Neural Networks in Novel Chromium-Based Alloy}, url = {https://doi.org/10.1039/d3cp00402c}, volume = {25}, year = {2023} } """ _DESCRIPTION = 'A comprehensive, two-tiered deep learning approach designed for precise object detection and segmentation in electron microscopy (EM) images.' _CATEGORIES = ["precipitate"] _HOMEPAGE = 'https://github.com/xiazeyu/DT_SegNet' _LICENSE = 'CC BY-NC 3.0' def convert_image(image_path): with open(image_path, "rb") as image_file: return image_file.read() # return Image.open(image_path) def convert_json(json_path): with open(json_path, "r") as json_file: json_str = json.dumps(json.load(json_file)) return json_str # .encode('utf-8') def convert_txt(txt_path): yolo_data = {"bbox": [], "category": []} # Open and read the text file with open(txt_path, "r") as file: for line in file: # Split each line into components parts = line.strip().split() # The first part is the category, which is added directly to the 'category' list yolo_data["category"].append(int(parts[0])) # The rest of the parts are the bounding box coordinates, which need to be converted to floats # and added as a sublist to the 'bbox' list bbox = [float(coord) for coord in parts[1:]] yolo_data["bbox"].append(bbox) return yolo_data def get_ds(pfx): image_array = [] seg_annotation_array = [] raw_seg_annotation_array = [] det_annotation_array = [] for img_idx in ids_split[pfx]: ydt = convert_txt(f"{pfx}/{img_idx}_label.txt") det_annotation_array.append({ "bbox": ydt["bbox"], "category": ydt["category"], }) image_array.append(convert_image(f"{pfx}/{img_idx}.png")) seg_annotation_array.append(convert_image(f"{pfx}/{img_idx}_label.png")) raw_seg_annotation_array.append(convert_json(f"{pfx}/{img_idx}.json")) data = { "id": ids_split[pfx], "original_name": [id_to_original[file] for file in ids_split[pfx]], "image": image_array, "det_annotation": det_annotation_array, "seg_annotation": seg_annotation_array, "raw_seg_annotation": raw_seg_annotation_array, } df = pd.DataFrame(data) features = datasets.Features({ 'id': datasets.Value('int8'), 'original_name': datasets.Value('string'), 'image': datasets.Image(), "det_annotation": datasets.Sequence( { "bbox": datasets.Sequence(datasets.Value("float32"), length=4), "category": datasets.ClassLabel(num_classes=1, names=_CATEGORIES), } ), 'seg_annotation': datasets.Image(), 'raw_seg_annotation': datasets.Value(dtype='string'), }) data_info = datasets.DatasetInfo( description=_DESCRIPTION, features=features, homepage=_HOMEPAGE, license=_LICENSE, citation=_CITATION, ) ds = datasets.Dataset.from_pandas(df, features=features, info=data_info, split=pfx) ds.VERSION = datasets.Version("1.0.0") return ds ddd = datasets.DatasetDict( { str(datasets.Split.TRAIN): get_ds(datasets.Split.TRAIN), str(datasets.Split.VALIDATION): get_ds(datasets.Split.VALIDATION), str(datasets.Split.TEST): get_ds(datasets.Split.TEST), } ) # ddd.save_to_disk('data/') # ddd.push_to_hub('xiazeyu/DT_SegNet')