COCO

README.md

@@ -26,6 +26,7 @@ This repository have done:
 - [x] YOLOv3 Head
 - [x] Keras Callbacks for Online Evaluation
 - [x] Load Official Weight File
+- [x] Data Format Converter(COCO and Pascal VOC)
 - [x] K-Means for Anchors
 - [x] Fight with 'NaN'
 - [x] Train (Strategy and Model Config)
@@ -63,7 +64,7 @@ path/to/image2 x1,y1,x2,y2,label
 ...
 ```
 
-Convert your data format firstly. We present a script for Pascal VOC in https://github.com/yuto3o/yolox/blob/master/data/pascal_voc/voc_convert.py
+Convert your data format firstly. We present [a script for Pascal VOC](./data/pascal_voc/voc_convert.py) and [a script for COCO](./data/coco/coco_convert.py).
 
 More details and a simple dataset could be got from https://github.com/YunYang1994/yymnist.
 

README_CN.md

@@ -14,7 +14,7 @@ YOLOv4-tiny（YOLOv4未提出，非官方）
 
 ## 被训练支配的恐惧 !
 
-- 因为目前可用的卡是一张游戏卡RzaiTX 2070S(8 G)，因此在训练时使用了较小的batch size。
+- 因为目前可用的卡是一张游戏卡RzaiTX 2070S(8 G)，因此在训练时使用了较小的batch，实际中尽量大batch可以省很多事。
 - 本项目的数据增强均使用在线形式，高级的数据增强方式会大大拖慢训练速度。
 - 训练过程中，Tiny版问题不大，而完整版模型容易NaN或者收敛慢，还在调参中。
 
@@ -26,6 +26,7 @@ This repository have done:
 - [x] YOLOv3 Head
 - [x] Keras Callbacks for Online Evaluation
 - [x] Load Official Weight File
+- [x] Data Format Converter(COCO and Pascal VOC)
 - [x] K-Means for Anchors
 - [x] Fight with 'NaN'
 - [x] Train (Strategy and Model Config)
@@ -63,7 +64,7 @@ path/to/image2 x1,y1,x2,y2,label
 ...
 ```
 
-当然本项目也提供了一个简单的VOC格式[转换脚本](./data/pascal_voc/voc_convert.py)。
+当然本项目也提供了一个的[VOC格式转换脚本](./data/pascal_voc/voc_convert.py)和一个[COCO格式转换脚本](./data/coco/coco_convert.py)。
 
 也可以从其他大佬的项目中看到这种格式的运用，甚至可以得到一个简单的入门级目标检测数据集： https://github.com/YunYang1994/yymnist.
 

data/coco/coco_convert.py

@@ -0,0 +1,99 @@
+# -*- coding: utf-8 -*-
+import os
+import json
+import pprint
+
+from absl import logging, app, flags
+from collections import defaultdict
+
+flags.DEFINE_string('coco_path', None, 'path to coco dataset')
+flags.DEFINE_string('name_path', None, 'path to coco name file')
+flags.DEFINE_string('txt_output_path', None, 'path to output txt file')
+flags.DEFINE_boolean('use_crowd', True, 'use crowd annotation')
+
+FLAGS = flags.FLAGS
+
+
+def convert(coco_path, coco_name_path, txt_output_path, use_crowd=True):
+    def _read_txt_line(path):
+        with open(path, 'r') as f:
+            txt = f.readlines()
+
+        return [line.strip() for line in txt]
+
+    ann_path_train2017 = os.path.join(coco_path, 'annotations', 'instances_train2017.json')
+    img_path_train2017 = os.path.join(coco_path, 'images', 'train2017')
+
+    ann_path_val2017 = os.path.join(coco_path, 'annotations', 'instances_val2017.json')
+    img_path_val2017 = os.path.join(coco_path, 'images', 'val2017')
+
+    coco_name = _read_txt_line(coco_name_path)
+
+    def _check_bbox(x1, y1, x2, y2, w, h):
+
+        if x1 < 0 or x2 < 0 or x1 > w or x2 > w or y1 < 0 or y2 < 0 or y1 > h or y2 > h:
+            logging.warning('cross boundary (' + str(w) + ',' + str(h) + '),(' + ','.join(
+                [str(x1), str(y1), str(x2), str(y2)]) + ')')
+
+            return str(min(max(x1, 0.), w)), str(min(max(y1, 0.), h)), str(min(max(x2, 0.), w)), str(
+                min(max(y2, 0.), h))
+
+        return x1, y1, x2, y2
+
+    def _write_to_text(ann_path, img_path, txt_path):
+        dataset = json.load(open(ann_path, 'r'))
+        print('creating index...')
+        anns, cats, imgs = {}, {}, {}
+        imgToAnns = defaultdict(list)
+        if 'annotations' in dataset:
+            for ann in dataset['annotations']:
+                imgToAnns[ann['image_id']].append(ann)
+                anns[ann['id']] = ann
+
+        if 'images' in dataset:
+            for img in dataset['images']:
+                imgs[img['id']] = img
+
+        if 'categories' in dataset:
+            for cat in dataset['categories']:
+                cats[cat['id']] = cat
+
+        print('Categories')
+        pprint.pprint(cats)
+        print('index created!')
+
+        with open(txt_path, 'w') as f:
+            for img_id, img in imgs.items():
+                anns = imgToAnns[img_id]
+                iw, ih = img['width'], img['height']
+                file_name = img['file_name']
+
+                line = os.path.join(img_path, file_name)
+
+                for ann in anns:
+
+                    label = cats[ann['category_id']]['name']
+                    if label not in coco_name:
+                        continue
+                    if not use_crowd and ann['iscrowd'] == 1:
+                        continue
+                    idx = coco_name.index(label)
+                    x, y, w, h = ann['bbox']
+                    x1, y1, x2, y2 = x, y, x + w, y + h
+                    x1, y1, x2, y2 = _check_bbox(x1, y1, x2, y2, iw, ih)
+
+                    line += ' ' + ','.join([str(x1), str(y1), str(x2), str(y2), str(idx)])
+
+                logging.info(line)
+                f.write(line + '\n')
+
+    _write_to_text(ann_path_train2017, img_path_train2017, os.path.join(txt_output_path, 'train2017.txt'))
+    _write_to_text(ann_path_val2017, img_path_val2017, os.path.join(txt_output_path, 'val2017.txt'))
+
+
+def main(_argv):
+    convert(FLAGS.coco_path, FLAGS.name_path, FLAGS.txt_output_path, FLAGS.use_crowd)
+
+
+if __name__ == '__main__':
+    app.run(main)