使用DSFD檢測DarkFace數(shù)據(jù)集過程

1.下載Dark Face數(shù)據(jù)集，使用track2.2_test_sample文件中圖片進(jìn)行人臉檢測測試。

2.修改DSFD源碼中demo.py部分：

test_oneimage()：
def test_oneimage():
# load net

# 影響網(wǎng)絡(luò)的自動求導(dǎo)機(jī)制，使網(wǎng)絡(luò)前向傳播后不進(jìn)行求導(dǎo)和反向傳播（僅測試時使用）
torch.set_grad_enabled(False)

# 加載config配置參數(shù)
cfg = widerface_640
# 分類的類別數(shù)目---widerface.py
num_classes = len(WIDERFace_CLASSES) + 1 # +1 background
# 加載SSD網(wǎng)絡(luò)模型，返回一個SSD實(shí)例
net = build_ssd('test', cfg['min_dim'], num_classes) # initialize SSD
# 加載預(yù)訓(xùn)練模型train_model
net.load_state_dict(torch.load(args.trained_model))
net.cuda()
# 表示進(jìn)入評估模式，神經(jīng)網(wǎng)絡(luò)中有train(),eval()兩種模式，使用eval()可關(guān)閉dropout
net.eval()
print('Finished loading model!')

# evaluation
cuda = args.cuda
transform = TestBaseTransform((104, 117, 123))
thresh=cfg['conf_thresh']
#save_path = args.save_folder
#num_images = len(testset)

# load data，從指定路徑加載待測圖像

'''
以此為界，前半部分為網(wǎng)絡(luò)模型加載和初始化，后半部分為單張圖片的人臉檢測。此處為測試圖片的路
徑設(shè)置過程。
修改部分，不適用arg.imag_root作為測試圖像路徑.
遍歷darkface數(shù)據(jù)集100張圖片，依次讀取并進(jìn)行測試
'''

folder = './data/'

#img_id = 'face'
for i in range(100):

img = cv2.imread(folder + str(i) + '_fake_B.jpg', cv2.IMREAD_COLOR)
img_id = 'test' + str(i)


# 單張圖片的測試過程
max_im_shrink = ( (2000.0*2000.0) / (img.shape[0] * img.shape[1])) ** 0.5
shrink = max_im_shrink if max_im_shrink < 1 else 1

det0 = infer(net , img , transform , thresh , cuda , shrink)
det1 = infer_flip(net , img , transform , thresh , cuda , shrink)
# shrink detecting and shrink only detect big face
st = 0.5 if max_im_shrink >= 0.75 else 0.5 * max_im_shrink
det_s = infer(net , img , transform , thresh , cuda , st)
index = np.where(np.maximum(det_s[:, 2] - det_s[:, 0] + 1, det_s[:, 3] - det_s[:, 1] + 1) > 30)[0]
det_s = det_s[index, :]
# enlarge one times
factor = 2
bt = min(factor, max_im_shrink) if max_im_shrink > 1 else (st + max_im_shrink) / 2
det_b = infer(net , img , transform , thresh , cuda , bt)
# enlarge small iamge x times for small face
if max_im_shrink > factor:
bt *= factor
while bt < max_im_shrink:
det_b = np.row_stack((det_b, infer(net , img , transform , thresh , cuda , bt)))
bt *= factor
det_b = np.row_stack((det_b, infer(net , img , transform , thresh , cuda , max_im_shrink) ))
# enlarge only detect small face
if bt > 1:
index = np.where(np.minimum(det_b[:, 2] - det_b[:, 0] + 1, det_b[:, 3] - det_b[:, 1] + 1) < 100)[0]
det_b = det_b[index, :]
else:
index = np.where(np.maximum(det_b[:, 2] - det_b[:, 0] + 1, det_b[:, 3] - det_b[:, 1] + 1) > 30)[0]
det_b = det_b[index, :]
det = np.row_stack((det0, det1, det_s, det_b))
det = bbox_vote(det)
vis_detections(img , det , img_id, args.visual_threshold)
3.運(yùn)行demo.py即可。能在arg.save_folder處得到100張dark_face人臉檢測結(jié)果。

補(bǔ)充：

DSFD只接收輸入格式為jpg的圖片，因此對darkface數(shù)據(jù)集進(jìn)行批量轉(zhuǎn)換。

import os
from PIL import Image

dirname_read="/home/...DSFD/darkface_png/"   # png格式圖片的輸入路徑
dirname_write="/home/...DSFD/data/"    # jpg圖片的輸出路徑
names=os.listdir(dirname_read)
count=0
for name in names:
   img=Image.open(dirname_read+name)
   name=name.split(".")
   if name[-1] == "png":
       name[-1] = "jpg"
       name = str.join(".", name)
       #r,g,b,a=img.split()
       #img=Image.merge("RGB",(r,g,b))
       to_save_path = dirname_write + name
       img.save(to_save_path)
       count+=1
       print(to_save_path, "------conut：", count)
   else:
       conti

審核編輯：湯梓紅