def transform_img(img_name, bbox):
if img_name[:4] == 'http':
type_ = 'url'
else:
type_ = 'path'
img = cv_load_image(img_name, type_) #.astype(np.float32)
# print bbox
# print len(bbox)
# raise
bbox = np.array(bbox).reshape((1, -1))
bbox = scale_bbox(bbox, 1.1).astype(int)[0]
H, W, _ = img.shape
bbox[2] = min(bbox[2], W)
bbox[3] = min(bbox[3], H)
img = img[bbox[1]:bbox[3], bbox[0]:bbox[2]]
img = cv2.resize(img, (224, 224))
#cv2.imshow('img',img)
#cv2.waitKey(0)
#hsv = img
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
hsv = hsv.transpose((2, 0, 1))
#print hsv.shape
#mean = np.array([104.0, 117.0, 123.0], dtype=np.float32).reshape((1, 1, -1))
#img -= mean
#img = img.transpose((2, 0, 1))
return hsv
def transform_img(img_name, bbox):
if img_name[:4] == 'http':
type_ = 'url'
else:
type_ = 'path'
img = cv_load_image(img_name, type_).astype(np.float32)
#img = cv_load_image(img_name, type_)
# print bbox
# print len(bbox)
# raise
bbox = np.array(bbox).reshape((1, -1))
bbox = scale_bbox(bbox, 1.1).astype(int)[0]
H, W, _ = img.shape
bbox[2] = min(bbox[2], W)
bbox[3] = min(bbox[3], H)
img = img[bbox[1]:bbox[3], bbox[0]:bbox[2]]
img = cv2.resize(img, (224, 224))
#plt.subplot(211)
#plt.imshow(img[:,:,(2,1,0)])
img = 1.2*img
img[img>255] = 255
#plt.subplot(212)
#plt.imshow(img[:,:,(2,1,0)])
#plt.show()
mean = np.array([104.0, 117.0, 123.0], dtype=np.float32).reshape((1, 1, -1))
img -= mean
img = img.transpose((2, 0, 1))
return img
def prep_im_for_blob_with_bbox(im, bbox, BBOX_SCALE):
im_old = im.astype(np.float32, copy=False)
bbox_old = bbox[:]
im = im.astype(np.float32, copy=False)
if config.BBOX_SCALE_TYPE == 'SCALE':
bbox = np.array(bbox).reshape((1, -1))
bbox = scale_bbox(bbox, config.BBOX_SCALE).astype(int)[0]
elif config.BBOX_SCALE_TYPE == 'ABSOLUTE':
bbox[0] -= int(BBOX_SCALE)
bbox[1] -= int(BBOX_SCALE)
bbox[2] += int(BBOX_SCALE)
bbox[3] += int(BBOX_SCALE)
else:
raise
H, W, _ = im.shape
bbox[0] = max(bbox[0], 0)
bbox[1] = max(bbox[1], 0)
bbox[2] = min(bbox[2], W)
bbox[3] = min(bbox[3], H)
im = im[bbox[1]:bbox[3], bbox[0]:bbox[2]]
target_size = config.TARGET_SIZE
pixel_means = np.array([[config.PIXEL_MEANS]])
if im.shape[0] * im.shape[1] == 0:
print im_old.shape, im.shape
print bbox_old, bbox
im = cv2.resize(im, (target_size, target_size),
interpolation=cv2.INTER_LINEAR)
im -= pixel_means
return im
def transform_img(img_name, bbox):
if img_name[:4] == 'http':
type_ = 'url'
else:
type_ = 'path'
img = cv_load_image(img_name, type_) #.astype(np.float32)
#plt.subplot(236), plt.imshow(img[:,:,(2,1,0)])
# print bbox
# print len(bbox)
# raise
bbox = np.array(bbox).reshape((1, -1))
bbox = scale_bbox(bbox, 1.1).astype(int)[0]
H, W, _ = img.shape
bbox[2] = min(bbox[2], W)
bbox[3] = min(bbox[3], H)
img = img[bbox[1]:bbox[3], bbox[0]:bbox[2]]
#plt.subplot(234), plt.imshow(img)
#plt.subplot(235), plt.imshow(img[:,:,(2,1,0)])
#img = cv2.resize(img, (224, 224))
#cv2.imshow('img',img)
#cv2.waitKey(0)
#hsv = img
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
#H, S, V = cv2.split(HSV)
#print hsv
#print hsv.shape
hist = [cv2.calcHist([hsv], [0], None, [50], [0, 180]), \
cv2.calcHist([hsv], [1], None, [50], [0, 256]), \
cv2.calcHist([hsv], [2], None, [50], [0, 256])]
hist = np.array(hist)
x = np.arange(50) + 0.5
#plt.subplot(231), plt.bar(x, hist[0])
#plt.subplot(232), plt.bar(x, hist[1])
#plt.subplot(233), plt.bar(x, hist[2])
#plt.bar(x, hist[0])
#plt.savefig('img/hist.jpg')
#hsv = hsv.transpose((2, 0, 1))
#mean = np.array([104.0, 117.0, 123.0], dtype=np.float32).reshape((1, 1, -1))
#img -= mean
#img = img.transpose((2, 0, 1))
return img, hsv, hist