updates

SmallObjectAugmentation/Helpers.py

@@ -0,0 +1,38 @@
+import glob
+import cv2 as cv2
+import numpy as np
+import matplotlib.pyplot as plt
+# import random
+import math
+from tqdm import tqdm
+
+
+def load_images(path):
+    image_list = []
+    images = glob.glob(path)
+    for index in range(len(images)):
+        image = cv2.cvtColor(cv2.imread(images[index]), cv2.COLOR_BGR2RGB)
+        image_list.append(image)
+        # image_list.append(cv2.resize(image,(1280,720)))
+
+    return image_list
+
+
+def read_images(path):
+    images = glob.glob(path)
+    return images
+
+
+def load_images_from_path(path):
+    image_list = []
+    for p in tqdm(path):
+        image = cv2.cvtColor(cv2.imread(p), cv2.COLOR_BGR2RGB)
+        image_list.append(image)
+    return image_list
+
+
+def replace_labels(path):
+    labelpath = []
+    for p in path:
+        labelpath.append(p.replace('.jpg', '.txt'))
+    return labelpath

SmallObjectAugmentation/README.md

@@ -0,0 +1,11 @@
+Simple python implementation of Augmentation for small object detection, as described in [Augmentation for small object detection
+](https://arxiv.org/pdf/1902.07296.pdf).
+
+![img1](https://github.com/gmayday1997/SmallObjectAugmentation/blob/master/img/2018-11-03_07_45_09_1.jpg)
+![img2](https://github.com/gmayday1997/SmallObjectAugmentation/blob/master/figs/2018-11-03_07_45_09_1_augment.jpg)
+
+## Demo
+```shell
+cd $ROOT
+python demo.py
+```

SmallObjectAugmentation/aug.py

@@ -0,0 +1,157 @@
+# import glob
+import cv2 as cv2
+import numpy as np
+# from PIL import Image
+import random
+import math
+from os.path import basename, split, join, dirname
+from util import *
+
+
+def find_str(filename):
+    if 'train' in filename:
+        return dirname(filename[filename.find('train'):])
+    else:
+        return dirname(filename[filename.find('val'):])
+
+
+def convert_all_boxes(shape, anno_infos, yolo_label_txt_dir):
+    height, width, n = shape
+    label_file = open(yolo_label_txt_dir, 'w')
+    for anno_info in anno_infos:
+        target_id, x1, y1, x2, y2 = anno_info
+        b = (float(x1), float(x2), float(y1), float(y2))
+        bb = convert((width, height), b)
+        label_file.write(str(target_id) + " " + " ".join([str(a) for a in bb]) + '\n')
+
+
+def save_crop_image(save_crop_base_dir, image_dir, idx, roi):
+    crop_save_dir = join(save_crop_base_dir, find_str(image_dir))
+    check_dir(crop_save_dir)
+    crop_img_save_dir = join(crop_save_dir, basename(image_dir)[:-3] + '_crop_' + str(idx) + '.jpg')
+    cv2.imwrite(crop_img_save_dir, roi)
+
+
+def copysmallobjects(image_dir, label_dir, save_base_dir, save_crop_base_dir=None,
+                     save_annoation_base_dir=None):
+    image = cv2.imread(image_dir)
+
+    labels = read_label_txt(label_dir)
+    if len(labels) == 0: return
+    rescale_labels = rescale_yolo_labels(labels, image.shape)  # 转换坐标表示
+    all_boxes = []
+
+    for idx, rescale_label in enumerate(rescale_labels):
+
+        all_boxes.append(rescale_label)
+        # 目标的长宽
+        rescale_label_height, rescale_label_width = rescale_label[4] - rescale_label[2], rescale_label[3] - \
+                                                    rescale_label[1]
+
+        if (issmallobject((rescale_label_height, rescale_label_width), thresh=64 * 64) and rescale_label[0] == '1'):
+            roi = image[rescale_label[2]:rescale_label[4], rescale_label[1]:rescale_label[3]]
+
+            new_bboxes = random_add_patches(rescale_label, rescale_labels, image.shape, paste_number=2, iou_thresh=0.2)
+            count = 0
+
+            # 将新生成的位置加入到label,并在相应位置画出物体
+            for new_bbox in new_bboxes:
+                count += 1
+                all_boxes.append(new_bbox)
+                cl, bbox_left, bbox_top, bbox_right, bbox_bottom = new_bbox[0], new_bbox[1], new_bbox[2], new_bbox[3], \
+                                                                   new_bbox[4]
+                try:
+                    if (count > 1):
+                        roi = flip_bbox(roi)
+                    image[bbox_top:bbox_bottom, bbox_left:bbox_right] = roi
+                except ValueError:
+                    continue
+
+    dir_name = find_str(image_dir)
+    save_dir = join(save_base_dir, dir_name)
+    check_dir(save_dir)
+    yolo_txt_dir = join(save_dir, basename(image_dir.replace('.jpg', '_augment.txt')))
+    cv2.imwrite(join(save_dir, basename(image_dir).replace('.jpg', '_augment.jpg')), image)
+    convert_all_boxes(image.shape, all_boxes, yolo_txt_dir)
+
+
+def GaussianBlurImg(image):
+    # 高斯模糊
+    ran = random.randint(0, 9)
+    if ran % 2 == 1:
+        image = cv2.GaussianBlur(image, ksize=(ran, ran), sigmaX=0, sigmaY=0)
+    else:
+        pass
+    return image
+
+
+def suo_fang(image, area_max=2000, area_min=1000):
+    # 改变图片大小
+    height, width, channels = image.shape
+
+    while (height*width) > area_max:
+        image = cv2.resize(image, (int(width * 0.9),int(height * 0.9)))
+        height, width, channels = image.shape
+        height,width = int(height*0.9),int(width*0.9)
+
+    while (height*width) < area_min:
+        image = cv2.resize(image, (int(width * 1.1),int(height * 1.1)))
+        height, width, channels = image.shape
+        height,width = int(height*1.1),int(width*1.1)
+
+    return image
+
+
+def copysmallobjects2(image_dir, label_dir, save_base_dir, small_img_dir):
+    image = cv2.imread(image_dir)
+    labels = read_label_txt(label_dir)
+    if len(labels) == 0:
+        return
+    rescale_labels = rescale_yolo_labels(labels, image.shape)  # 转换坐标表示
+    all_boxes = []
+    for _, rescale_label in enumerate(rescale_labels):
+        all_boxes.append(rescale_label)
+
+    for small_img_dirs in small_img_dir:
+        image_bbox = cv2.imread(small_img_dirs)
+        #roi = image_bbox
+        roi = suo_fang(image_bbox,area_max=3000,area_min=1500)
+
+        new_bboxes = random_add_patches2(roi.shape, rescale_labels, image.shape, paste_number=1, iou_thresh=0)
+        count = 0
+        for new_bbox in new_bboxes:
+            count += 1
+
+            cl, bbox_left, bbox_top, bbox_right, bbox_bottom = new_bbox[0], new_bbox[1], new_bbox[2], new_bbox[3], \
+                                                               new_bbox[4]
+            #roi = GaussianBlurImg(roi)  # 高斯模糊
+            height, width, channels = roi.shape
+            center = (int(width / 2),int(height / 2))
+            #ran_point = (int((bbox_top+bbox_bottom)/2),int((bbox_left+bbox_right)/2))
+            mask = 255 * np.ones(roi.shape, roi.dtype)
+
+            try:
+                if count > 1:
+                    roi = flip_bbox(roi)
+                #image[bbox_top:bbox_bottom, bbox_left:bbox_right] = roi
+                #image[bbox_top:bbox_bottom, bbox_left:bbox_right] = cv2.addWeighted(image[bbox_top:bbox_bottom, bbox_left:bbox_right],
+                #                                                                    0.5,roi,0.5,0) #图片融合
+
+                # 泊松融合
+                #image = cv2.seamlessClone(roi, image, mask, ran_point, cv2.NORMAL_CLONE)
+                #print(str(bbox_bottom-bbox_top) + "|" + str(bbox_right-bbox_left))
+                #print(roi.shape)
+                #print(mask.shape)
+                image[bbox_top:bbox_bottom, bbox_left:bbox_right] = cv2.seamlessClone(roi, image[bbox_top:bbox_bottom, bbox_left:bbox_right],
+                                                                                      mask, center, cv2.NORMAL_CLONE)
+                all_boxes.append(new_bbox)
+                rescale_labels.append(new_bbox)
+            except ValueError:
+                print("---")
+                continue
+    dir_name = find_str(image_dir)
+    save_dir = join(save_base_dir, dir_name)
+    check_dir(save_dir)
+    yolo_txt_dir = join(save_dir, basename(image_dir.replace('.jpg', '_augment.txt')))
+    cv2.imwrite(join(save_dir, basename(image_dir).replace('.jpg', '_augment.jpg')), image)
+    convert_all_boxes(image.shape, all_boxes, yolo_txt_dir)

SmallObjectAugmentation/background/img00000.txt

@@ -0,0 +1,3 @@
+0 0.265364583333 0.139130434783 0.0151041666667 0.0369565217391
+0 0.284114583333 0.151086956522 0.0119791666667 0.0173913043478
+0 0.3515625 0.179347826087 0.015625 0.0239130434783

SmallObjectAugmentation/background/img00001.txt

@@ -0,0 +1,3 @@
+0 0.2671875 0.139130434783 0.021875 0.0434782608696
+0 0.284114583333 0.154891304348 0.0119791666667 0.025
+0 0.350520833333 0.179347826087 0.0177083333333 0.0239130434783

SmallObjectAugmentation/break/img00000.txt

@@ -0,0 +1,3 @@
+0 0.265364583333 0.139130434783 0.0151041666667 0.0369565217391
+0 0.284114583333 0.151086956522 0.0119791666667 0.0173913043478
+0 0.3515625 0.179347826087 0.015625 0.0239130434783

SmallObjectAugmentation/break/img00001.txt

@@ -0,0 +1,3 @@
+0 0.2671875 0.139130434783 0.021875 0.0434782608696
+0 0.284114583333 0.154891304348 0.0119791666667 0.025
+0 0.350520833333 0.179347826087 0.0177083333333 0.0239130434783

SmallObjectAugmentation/createTrain.sh

@@ -0,0 +1,2 @@
+find ./background -maxdepth 4 -name "*.jpg" | sort > train.txt
+find ./crops -maxdepth 4 -name "*.jpg" | sort > small.txt

SmallObjectAugmentation/demo.py

@@ -0,0 +1,29 @@
+import aug as am
+import Helpers as hp
+from util import *
+import os
+from os.path import join
+from tqdm import tqdm
+import random
+
+base_dir = os.getcwd()
+
+save_base_dir = join(base_dir, 'save')
+
+check_dir(save_base_dir)
+
+imgs_dir = [f.strip() for f in open(join(base_dir, 'train.txt')).readlines()]
+labels_dir = hp.replace_labels(imgs_dir)
+
+small_imgs_dir = [f.strip() for f in open(join(base_dir, 'small.txt')).readlines()]
+random.shuffle(small_imgs_dir)
+
+for image_dir, label_dir in tqdm(zip(imgs_dir, labels_dir)):
+    small_img = []
+    for x in range(8):
+        if small_imgs_dir == []:
+            #exit()
+            small_imgs_dir = [f.strip() for f in open(join(base_dir,'small.txt')).readlines()]
+            random.shuffle(small_imgs_dir)
+        small_img.append(small_imgs_dir.pop())
+    am.copysmallobjects2(image_dir, label_dir, save_base_dir,small_img)

SmallObjectAugmentation/figs/2018-11-03_07_45_09_1_augment.txt

@@ -0,0 +1,7 @@
+1 0.6 0.5898148148148148 0.0125 0.061111111111111116
+1 0.346875 0.5916666666666667 0.0125 0.061111111111111116
+1 0.2708333333333333 0.5750000000000001 0.0125 0.061111111111111116
+0 0.3046875 0.44537037037037036 0.065625 0.1388888888888889
+0 0.10625 0.8481481481481482 0.19375 0.3
+0 0.6140625 0.40740740740740744 0.026041666666666668 0.037037037037037035
+0 0.6359375 0.375 0.036458333333333336 0.07222222222222223

SmallObjectAugmentation/img/2018-11-03_07_45_09_1.txt

@@ -0,0 +1,5 @@
+1 0.602083333333 0.591666666667 0.0125 0.0611111111111
+0 0.306770833333 0.449074074074 0.065625 0.138888888889
+0 0.108333333333 0.85 0.19375 0.3
+0 0.616145833333 0.411111111111 0.0260416666667 0.037037037037
+0 0.636979166667 0.376851851852 0.0364583333333 0.0722222222222

SmallObjectAugmentation/save/img00000_augment.txt

@@ -0,0 +1,13 @@
+0 0.2643229166666667 0.13804347826086957 0.015104166666666667 0.03695652173913044
+0 0.2830729166666667 0.15 0.011979166666666666 0.017391304347826087
+0 0.35104166666666664 0.1782608695652174 0.015625 0.02391304347826087
+1 0.8265625 0.6293478260869565 0.011458333333333333 0.06739130434782609
+1 0.5752604166666666 0.7494565217391305 0.021354166666666667 0.07934782608695652
+1 0.6140625 0.8445652173913044 0.014583333333333334 0.09565217391304348
+1 0.72890625 0.6347826086956522 0.015104166666666667 0.09130434782608696
+1 0.43515624999999997 0.7663043478260869 0.022395833333333334 0.0782608695652174
+1 0.7671875 0.5396739130434782 0.017708333333333333 0.0923913043478261
+1 0.5122395833333333 0.9032608695652175 0.0203125 0.08695652173913043
+1 0.53046875 0.5679347826086957 0.018229166666666668 0.08586956521739131
+1 0.6596354166666667 0.7711956521739131 0.021354166666666667 0.07282608695652174
+1 0.490625 0.8190217391304347 0.01875 0.08804347826086957

SmallObjectAugmentation/save/img00001_augment.txt

@@ -0,0 +1,13 @@
+0 0.26666666666666666 0.13804347826086957 0.021875 0.043478260869565216
+0 0.2830729166666667 0.15380434782608696 0.011979166666666666 0.025
+0 0.3494791666666667 0.1782608695652174 0.017708333333333333 0.02391304347826087
+1 0.94296875 0.6923913043478261 0.013020833333333334 0.07608695652173914
+1 0.47031249999999997 0.8630434782608696 0.015625 0.07173913043478261
+1 0.62265625 0.5782608695652174 0.024479166666666666 0.08478260869565217
+1 0.29921875 0.5173913043478261 0.0203125 0.07173913043478261
+1 0.7361979166666667 0.5119565217391304 0.013020833333333334 0.06304347826086956
+1 0.3963541666666667 0.6402173913043478 0.016666666666666666 0.1
+1 0.4236979166666667 0.7717391304347826 0.0171875 0.08260869565217391
+1 0.9356770833333333 0.8913043478260869 0.0203125 0.09782608695652174
+1 0.8783854166666667 0.5608695652173913 0.015104166666666667 0.09130434782608696
+1 0.8260416666666667 0.5233695652173913 0.019791666666666666 0.08369565217391305

SmallObjectAugmentation/small.txt

@@ -0,0 +1,20 @@
+./crops/0000003_crop_39.jpg
+./crops/0000003_crop_60.jpg
+./crops/0000003_crop_73.jpg
+./crops/0000003_crop_74.jpg
+./crops/0000003_crop_87.jpg
+./crops/0000004_crop_28.jpg
+./crops/0000004_crop_30.jpg
+./crops/0000004_crop_5.jpg
+./crops/0000004_crop_6.jpg
+./crops/0000004_crop_7.jpg
+./crops/0000004_crop_8.jpg
+./crops/0000005_crop_5.jpg
+./crops/0000014_crop_15.jpg
+./crops/0000024_crop_5.jpg
+./crops/0000188_crop_44.jpg
+./crops/0000189_crop_76.jpg
+./crops/0000189_crop_77.jpg
+./crops/0000190_crop_28.jpg
+./crops/0000239_crop_23.jpg
+./crops/0000242_crop_15.jpg

SmallObjectAugmentation/train.txt

@@ -0,0 +1,2 @@
+./background/img00000.jpg
+./background/img00001.jpg