def demo(sess, net, image_name,imagedir, mode,pedestrian,cyclist):
"""Detect object classes in an image using pre-computed object proposals."""
# Load the demo image
im_file = os.path.join(imagedir, image_name)
im = cv2.imread(im_file)
save_path='result_image_new'
# Detect all object classes and regress object bounds
if mode == 'fast':
scores, boxes = im_detect_fast(sess, net, im)
else:
multi_scales=[600,800,1000]
scores, boxes = im_detect(sess, net, im,multi_scales)
# Visualize detections for each class
CONF_THRESH = 0.4
# Visualize people
for cls_ind,cls in enumerate(CLASSES[1:]):
cls_ind += 1
# cls = 'pedestrian'
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=soft_nms(dets,sigma=0.6,Nt=0.3,method=2)
dets=keep
# if cls=='pedestrian':
# map_file=pedestrian
# elif cls=='cyclist':
map_file=pedestrian
vis_detections(im, image_name, cls, dets,map_file,thresh=CONF_THRESH)
def demo(sess, net, mode, im):
"""Detect object classes in an image using pre-computed object proposals."""
if mode == 'fast':
scores, boxes = im_detect_fast(sess, net, im)
else:
multi_scales = size
scores, boxes = im_detect(sess, net, im, multi_scales)
# Visualize detections for each class
CONF_THRESH = soft_thresh
# Visualize people
for cls_ind, cls in enumerate(CLASSES[1:]):
cls_ind += 1
# cls = 'pedestrian'
#image_name=image_name.split('/')[-1].split('.')[0]
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 = soft_nms(dets, sigma=0.6, Nt=0.3, method=2)
dets = keep
#print(dets)
inds = np.where(dets[:, -1] >= CONF_THRESH)[0]
bbox_list = []
for i in inds:
bbox = dets[i, :4]
score = np.round(dets[i, -1] * 1000) / 1000
xmin = round(bbox[0], 1)
ymin = round(bbox[1], 1)
xmax = round(bbox[2], 1)
ymax = round(bbox[3], 1)
bbox_list.append([xmin, ymin, xmax, ymax, score])
return bbox_list
def demo(sess, net, image_name,xminarr,yminarr,xmaxarr,ymaxarr,results,score_file,index_file,num_boxes,imagedir, mode):
"""Detect object classes in an image using pre-computed object proposals."""
# Load the demo image
im_file = os.path.join(imagedir, image_name)
im = cv2.imread(im_file)
# Detect all object classes and regress object bounds
if mode == 'fast':
scores, boxes = im_detect_fast(sess, net, im)
else:
scores, boxes = im_detect(sess, net, im)
# Visualize detections for each class
CONF_THRESH = 0.1
# Visualize people
cls_ind = 1
cls = 'person'
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=soft_nms(dets,method=2)
dets=keep
if(dets.shape[0]!=0):
index_file.write("{} {} ".format(image_name,num_boxes+1))
num_boxes = vis_detections(im, image_name, cls, dets,xminarr,yminarr,xmaxarr,ymaxarr,results,score_file,index_file,num_boxes, thresh=CONF_THRESH)
if(dets.shape[0]!=0):
index_file.write("{}\n".format(num_boxes))
return num_boxes
def demo(sess, net, image_name, imagedir, mode, map_file):
"""Detect object classes in an image using pre-computed object proposals."""
#print(image_name)
im = cv2.imread(image_name)
#print(im)
save_path = 'result_image_new'
# Detect all object classes and regress object bounds
if mode == 'fast':
scores, boxes = im_detect_fast(sess, net, im)
else:
multi_scales = size
scores, boxes = im_detect(sess, net, im, multi_scales)
# Visualize detections for each class
CONF_THRESH = soft_thresh
# Visualize people
for cls_ind, cls in enumerate(CLASSES[1:]):
cls_ind += 1
# cls = 'pedestrian'
image_name = image_name.split('/')[-1].split('.')[0]
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 = soft_nms(dets, sigma=0.6, Nt=0.3, method=2)
dets = keep
#print(dets)
vis_detections(im, image_name, cls, dets, map_file, thresh=CONF_THRESH)
def demo(sess, net, image_name, imagedir, mode):
"""Detect object classes in an image using pre-computed object proposals."""
# Load the demo image
im_file = os.path.join(imagedir, image_name)
im = cv2.imread(im_file)
save_path = 'wider_result_1class_image'
if not os.path.exists(save_path):
os.mkdir(save_path)
# Detect all object classes and regress object bounds
if mode == 'fast':
scores, boxes = im_detect_fast(sess, net, im)
else:
multi_scales = [700, 800, 900]
scores, boxes = im_detect(sess, net, im, multi_scales)
# Visualize detections for each class
CONF_THRESH = 0.4
# Visualize people
for cls_ind, cls in enumerate(CLASSES[1:]):
cls_ind += 1
# cls = 'pedestrian'
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 = soft_nms(dets, sigma=0.6, Nt=0.3, method=2)
dets = keep
vis_detections(im, image_name, cls, dets, thresh=CONF_THRESH)
image_name = image_name.split('/')[-1].split('.')[0] + '.jpg'
cv2.imwrite(os.path.join(save_path, image_name), im)
def demo(sess, net, image_name, xminarr, yminarr, xmaxarr, ymaxarr, results,
score_file, index_file, num_boxes):
"""Detect object classes in an image using pre-computed object proposals."""
# Load the demo image
# im_file = os.path.join(cfg.DATA_DIR, 'coco-minival_images', image_name)
# im_file = os.path.join(cfg.DATA_DIR, 'coco-test-dev', image_name)
im_file = os.path.join('/SSD2/jeff/coco', 'trainval2017', image_name)
im = cv2.imread(im_file)
# Detect all object classes and regress object bounds
timer = Timer()
timer.tic()
scores, boxes = im_detect(sess, net, im)
timer.toc()
print('Detection took {:.3f}s for {:d} object proposals'.format(
timer.total_time, boxes.shape[0]))
# Visualize detections for each class
CONF_THRESH = 0.035
NMS_THRESH = 0.7
# Visualize people
cls_ind = 1
cls = 'person'
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 = nms(dets, NMS_THRESH)
keep = soft_nms(dets, method=2)
#dets = dets[keep, :]
dets = keep
if (dets.shape[0] != 0):
index_file.write("{} {} ".format(image_name, num_boxes + 1))
num_boxes = vis_detections(im,
image_name,
cls,
dets,
xminarr,
yminarr,
xmaxarr,
ymaxarr,
results,
score_file,
index_file,
num_boxes,
thresh=CONF_THRESH)
if (dets.shape[0] != 0):
index_file.write("{}\n".format(num_boxes))
# for cls_ind, cls in enumerate(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 = nms(dets, NMS_THRESH)
# dets = dets[keep, :]
# vis_detections(im, cls, dets, thresh=CONF_THRESH)
return num_boxes
def demo(sess, net, image_name, xminarr, yminarr, xmaxarr, ymaxarr, results,
score_file, index_file, num_boxes, imagedir, mode):
"""Detect object classes in an image using pre-computed object proposals."""
# Load the demo image
im_file = os.path.join(imagedir, image_name)
im = cv2.imread(im_file)
##################
name_parts = image_name.split('/')
file_name = '/disks/data4/zyli/Faster-RCNN-AlphaPose/heatmap/%s/%s.npy' % (
name_parts[-2], name_parts[-1].replace('.jpg', ''))
assert (os.path.exists(file_name))
hm = np.array(np.load(file_name), np.float32)
hm = cv2.resize(hm, (im.shape[1], im.shape[0]),
interpolation=cv2.INTER_LINEAR)
im = np.concatenate([im, hm], axis=2)
##################
# Detect all object classes and regress object bounds
if mode == 'fast':
scores, boxes = im_detect_fast(sess, net, im)
else:
scores, boxes = im_detect(sess, net, im)
# Visualize detections for each class
CONF_THRESH = 0.1
# Visualize people
cls_ind = 1
cls = 'person'
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 = soft_nms(dets, method=2)
dets = keep
if (dets.shape[0] != 0):
index_file.write("{} {} ".format(image_name, num_boxes + 1))
num_boxes = vis_detections(im,
image_name,
cls,
dets,
xminarr,
yminarr,
xmaxarr,
ymaxarr,
results,
score_file,
index_file,
num_boxes,
thresh=CONF_THRESH)
if (dets.shape[0] != 0):
index_file.write("{}\n".format(num_boxes))
return num_boxes
def detect(sess, net, imagedir, mode):
"""Detects object classes in an image using pre-computed object proposals."""
# Load the demo image
im = cv2.imread(imagedir)
# Detect all object classes and regress object bounds
if mode == 'fast':
scores, boxes = im_detect_fast(sess, net, im)
else:
scores, boxes = im_detect(sess, net, im)
# boxes for people
cls_ind = 1
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)
dets = soft_nms(dets, method=2)
xminarr, yminarr, xmaxarr, ymaxarr = extract_bbox(dets)
return xminarr, yminarr, xmaxarr, ymaxarr
def demo(sess, net, image_name, xminarr, yminarr, xmaxarr, ymaxarr, results,
score_file, index_file, num_boxes, imagedir, mode, map_file):
"""Detect object classes in an image using pre-computed object proposals."""
# Load the demo image
im_file = os.path.join(imagedir, image_name)
im = cv2.imread(im_file)
# Detect all object classes and regress object bounds
if mode == 'fast':
scores, boxes = im_detect_fast(sess, net, im)
else:
scores, boxes = im_detect(sess, net, im)
# Visualize detections for each class
CONF_THRESH = 0.5
# Visualize people
cls_ind = 1
cls = 'person'
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 = soft_nms(dets, method=2)
dets = keep
if (dets.shape[0] != 0):
index_file.write("{} {} ".format(image_name, num_boxes + 1))
num_boxes = vis_detections(im,
image_name,
cls,
dets,
xminarr,
yminarr,
xmaxarr,
ymaxarr,
results,
score_file,
index_file,
num_boxes,
map_file,
thresh=CONF_THRESH)
if (dets.shape[0] != 0):
index_file.write("{}\n".format(num_boxes))
return num_boxes
filename = os.path.join(image_dir, img_name)
if os.path.exists(filename):
image = skimage.io.imread(filename)
[img_height, img_width, _] = image.shape
boxes, scores = detect(sess, net, image)
dets_ = np.hstack((boxes, scores[:,
np.newaxis])).astype(np.float32)
# prvs = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)
# next = cv2.cvtColor(image_mid,cv2.COLOR_BGR2GRAY)
# opflow = cv2.calcOpticalFlowFarneback(prvs,next, None, 0.5, 3, 15, 3, 5, 1.2, 0)
# dets = np.vstack((dets, op_argument(dets_, opflow, img_height, img_width)))
dets = np.vstack((dets, dets_))
keep = soft_nms(dets, method=2)
detections = keep
# print(detections.shape)
# Parse the outputs.
det_conf = detections[:, 4]
det_xmin = detections[:, 1]
det_ymin = detections[:, 0]
det_xmax = detections[:, 3]
det_ymax = detections[:, 2]
top_indices1 = [m for m, conf in enumerate(det_conf) if conf > 0.1]
top_indices3 = [
m for m, height in enumerate(det_ymax - det_ymin)
if height > 0.1 * img_height
]
def demo_seq(sess, net, imgname,xminarr,yminarr,xmaxarr,ymaxarr,results,score_file,index_file,num_boxes,imagedir, fast_mode):
"""Detect object classes in an image using pre-computed object proposals."""
idx = int(imgname.split('.')[0])
# Load the demo image
#im_file = os.path.join(cfg.DATA_DIR, 'coco-minival_images', image_name)
im_file = os.path.join(imagedir, '%05d.jpg'%(idx))
im = cv2.imread(im_file)
# Detect all object classes and regress object bounds
timer = Timer()
timer.tic()
if fast_mode == True:
scores, boxes = im_detect_fast(sess, net, im)
else:
scores, boxes = im_detect(sess, net, im)
# Visualize detections for each class
CONF_THRESH = 0.1
NMS_THRESH = 0.7
# Visualize people
cls_ind = 1
cls = 'person'
cls_boxes = boxes[:, 4*cls_ind:4*(cls_ind + 1)]
cls_scores = scores[:, cls_ind]
#Prev image
im_file = os.path.join(imagedir, '%05d.jpg'%(idx-1))
if os.path.isfile(im_file):
im = cv2.imread(im_file)
if fast_mode == True:
scores, boxes = im_detect_fast(sess, net, im)
else:
scores, boxes = im_detect(sess, net, im)
# Visualize people
cls_ind = 1
cls_boxes = np.append(cls_boxes,boxes[:, 4*cls_ind:4*(cls_ind + 1)], axis=0)
cls_scores = np.append(cls_scores,scores[:, cls_ind], axis=0)
#Prev Prev image
im_file = os.path.join(imagedir, '%05d.jpg'%(idx-2))
if os.path.isfile(im_file):
im = cv2.imread(im_file)
if fast_mode == True:
scores, boxes = im_detect_fast(sess, net, im)
else:
scores, boxes = im_detect(sess, net, im)
# Visualize people
cls_ind = 1
cls_boxes = np.append(cls_boxes,boxes[:, 4*cls_ind:4*(cls_ind + 1)], axis=0)
cls_scores = np.append(cls_scores,scores[:, cls_ind], axis=0)
#Next image
im_file = os.path.join(imagedir, '%05d.jpg'%(idx+1))
if os.path.isfile(im_file):
im = cv2.imread(im_file)
if fast_mode == True:
scores, boxes = im_detect_fast(sess, net, im)
else:
scores, boxes = im_detect(sess, net, im)
# Visualize people
cls_ind = 1
cls_boxes = np.append(cls_boxes,boxes[:, 4*cls_ind:4*(cls_ind + 1)], axis=0)
cls_scores = np.append(cls_scores,scores[:, cls_ind], axis=0)
#Next Next image
im_file = os.path.join(imagedir, '%05d.jpg'%(idx+2))
if os.path.isfile(im_file):
im = cv2.imread(im_file)
if fast_mode == True:
scores, boxes = im_detect_fast(sess, net, im)
else:
scores, boxes = im_detect(sess, net, im)
# Visualize people
cls_ind = 1
cls_boxes = np.append(cls_boxes,boxes[:, 4*cls_ind:4*(cls_ind + 1)], axis=0)
cls_scores = np.append(cls_scores,scores[:, cls_ind], axis=0)
timer.toc()
# print('Detection took {:.3f}s for {:d} object proposals'.format(timer.total_time, boxes.shape[0]))
dets = np.hstack((cls_boxes,
cls_scores[:, np.newaxis])).astype(np.float32)
#keep = nms(dets, NMS_THRESH)
keep=soft_nms(dets,method=2)
#dets = dets[keep, :]
dets=keep
if(dets.shape[0]!=0):
index_file.write("{} {} ".format('%05d.jpg'%(idx),num_boxes+1))
num_boxes = vis_detections(im, '%05d.jpg'%(idx), cls, dets,xminarr,yminarr,xmaxarr,ymaxarr,results,score_file,index_file,num_boxes, thresh=CONF_THRESH)
if(dets.shape[0]!=0):
index_file.write("{}\n".format(num_boxes))
# for cls_ind, cls in enumerate(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 = nms(dets, NMS_THRESH)
# dets = dets[keep, :]
# vis_detections(im, cls, dets, thresh=CONF_THRESH)
return num_boxes
