def process_image(filename):
print 'Process image: ', filename
#image = PIL.Image.open(filename)
if not os.path.isfile(filename):
print 'Image file %s not exist' % filename
return
image = D.cv2.imread(filename)
#data.set_data(image)
show_img = image[:, :, (2, 1, 0)]
ax.cla()
ax.imshow(show_img, aspect='equal')
print 'Start detection'
timer = D.Timer()
timer.tic()
scores, boxes = D.im_detect(net, image)
timer.toc()
print('Detection took {:.3f}s for '
'{:d} object proposals').format(timer.total_time, boxes.shape[0])
print 'Plot result.'
CONF_THRESH = 0.8
NMS_THRESH = 0.3
for cls_ind, cls in enumerate(D.CLASSES[1:]):
cls_ind += 1 # because we skipped background
cls_boxes = boxes[:, 4 * cls_ind:4 * (cls_ind + 1)]
cls_scores = scores[:, cls_ind]
dets = np.hstack(
(cls_boxes, cls_scores[:, np.newaxis])).astype(np.float32)
keep = D.nms(dets, NMS_THRESH)
dets = dets[keep, :]
plot_rect(image, cls, dets, thresh=CONF_THRESH)
fig.canvas.draw()
def faster_rcnn_detect(ax, image):
if not net:
init_caffe() # Caffe needs to be started in this thread, otherwise GIL will make it very slow
r,g,b,a = cv2.split(image)
bgr_img = cv2.merge([b,g,r])
timer = D.Timer()
timer.tic()
scores, boxes = D.im_detect(net, bgr_img)
timer.toc()
print ('Detection took {:.3f}s for '
'{:d} object proposals').format(timer.total_time, boxes.shape[0])
plot_image(ax, image, boxes, scores)
return image
def faster_rcnn_detect(ax, image):
if not net:
init_caffe(
) # Caffe needs to be started in this thread, otherwise GIL will make it very slow
r, g, b, a = cv2.split(image)
bgr_img = cv2.merge([b, g, r])
timer = D.Timer()
timer.tic()
scores, boxes = D.im_detect(net, bgr_img)
timer.toc()
print(('Detection took {:.3f}s for '
'{:d} object proposals').format(timer.total_time, boxes.shape[0]))
plot_image(ax, image, boxes, scores)
return image
def init_caffe(): # TODO: parse args into here
global net
prototxt = os.path.join(D.cfg.MODELS_DIR, D.NETS['vgg16'][0],
'faster_rcnn_alt_opt', 'faster_rcnn_test.pt')
caffemodel = os.path.join(D.cfg.DATA_DIR, 'faster_rcnn_models'
, D.NETS['vgg16'][1])
gpu_id = 0
D.caffe.set_mode_gpu()
D.caffe.set_device(gpu_id)
D.cfg.GPU_ID = gpu_id
D.cfg.TEST.HAS_RPN = True
net = D.caffe.Net(prototxt, caffemodel, D.caffe.TEST)
# Warmup on a dummy image
im = 128 * np.ones((300, 500, 3), dtype = np.uint8)
for _ in xrange(2):
_, _ = D.im_detect(net, im)
def init_caffe(): # TODO: parse args into here
global net
prototxt = os.path.join(D.cfg.MODELS_DIR, D.NETS['vgg16'][0],
'faster_rcnn_alt_opt', 'faster_rcnn_test.pt')
caffemodel = os.path.join(D.cfg.DATA_DIR, 'faster_rcnn_models',
D.NETS['vgg16'][1])
gpu_id = 0
D.caffe.set_mode_gpu()
D.caffe.set_device(gpu_id)
D.cfg.GPU_ID = gpu_id
D.cfg.TEST.HAS_RPN = True
net = D.caffe.Net(prototxt, caffemodel, D.caffe.TEST)
# Warmup on a dummy image
im = 128 * np.ones((300, 500, 3), dtype=np.uint8)
for _ in range(2):
_, _ = D.im_detect(net, im)
def process_image(filename):
if not net:
init_caffe(
) # Caffe needs to be started in this thread, otherwise GIL will make it very slow
print 'Process image: %s' % filename
if not os.path.isfile(filename):
print 'Image file %s not exist' % filename
return
image = D.cv2.imread(filename)
timer = D.Timer()
timer.tic()
scores, boxes = D.im_detect(net, image)
timer.toc()
print('Detection took {:.3f}s for '
'{:d} object proposals').format(timer.total_time, boxes.shape[0])
show_img = image[:, :, (2, 1, 0)] # Reorder to RGB
plot_image(show_img, boxes, scores)