def nms(dets, thresh, force_cpu=False):
"""Dispatch to either CPU or GPU NMS implementations."""
if dets.shape[0] == 0:
return []
if cfg.USE_GPU_NMS and not force_cpu:
return py_cpu_nms(dets, thresh)
else:
return py_cpu_nms(dets, thresh)
def process_image(net, im_file, outfolder, prob_threshold=0.5, nms_threshold=0.3, imgprefix=""):
"""Detect object classes in an image using pre-computed object proposals."""
# Load the demo image
im = cv2.imread(im_file)
# Detect all object classes and regress object bounds
timer = Timer()
timer.tic()
scores, boxes = im_detect(net, im)
timer.toc()
print("Detection took {:.3f}s for {:d} object proposals".format(timer.total_time, boxes.shape[0]))
(dir, name) = os.path.split(im_file)
imgname = os.path.join(dir, imgprefix + name)
# Visualize detections for each class
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)
# TODO: for some reason native nms is not working properly
# I have replaced it by the python version by now, but
# idea in the future would it be to solve it and uncomment the
# the following line and comment the python nms version.
# keep = nms(dets, nms_threshold)
keep = py_cpu_nms(dets, nms_threshold)
dets = dets[keep, :]
save_detections(im, cls, dets, imgname, outfolder, thresh=prob_threshold)
def apply_nms(dets, nms_thresh):
from nms.py_cpu_nms import py_cpu_nms
# Group detections into matrix form:
# [[top, left, bottom, right, score], ...]
per_class_dets = dict()
for det_i, det in enumerate(dets):
if det.cls not in per_class_dets:
per_class_dets[det.cls] = [[
det.top, det.left, det.bottom, det.right, det.score, det_i
]]
else:
per_class_dets[det.cls].append(
[det.top, det.left, det.bottom, det.right, det.score, det_i])
# Apply NMS
class_removals = dict()
idx_det_keep = []
for k in per_class_dets.keys():
per_class_dets[k] = np.array(per_class_dets[k])
idx_nms_keep = py_cpu_nms(per_class_dets[k], nms_thresh)
if len(idx_nms_keep) < len(per_class_dets[k]):
class_removals[k] = len(per_class_dets[k]) - len(idx_nms_keep)
idx_det_keep += list(per_class_dets[k][idx_nms_keep, -1])
filtered_dets = []
for idx in idx_det_keep:
filtered_dets.append(dets[int(idx)])
return (filtered_dets, class_removals)
def demo(net, image_name):
"""Detect object classes in an image using pre-computed object proposals."""
global ax
# Load the demo image
im_file = os.path.join(cfg.DATA_DIR, 'demo', image_name)
im = cv2.imread(im_file)
# Detect all object classes and regress object bounds
timer = Timer()
timer.tic()
scores, boxes = im_detect(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 = 0.6
NMS_THRESH = 0.3
im = im[:, :, (2, 1, 0)]
fig, ax = plt.subplots(figsize=(6.4, 4.8))
ax.imshow(im, aspect='equal')
animated = False
all_dets = []
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]
cls_inds = np.ones((len(cls_scores), 1)) * cls_ind
# col 4 and col -1 are both cls_scores, to satisfy nms()
dets = np.hstack((cls_boxes, cls_scores[:, np.newaxis], cls_inds,
cls_scores[:, np.newaxis])).astype(np.float32)
all_dets.append(dets)
all_dets = np.concatenate(all_dets)
conf_inds = all_dets[:, 4] >= CONF_THRESH
conf_dets = all_dets[conf_inds]
keep = py_cpu_nms(conf_dets, NMS_THRESH) # nms(conf_dets, NMS_THRESH) #
nms_dets = conf_dets[keep, :]
vis_detections(im, ax, nms_dets, thresh=CONF_THRESH, animated=animated)
plt.axis('off')
plt.tight_layout()
plt.title(image_name)
if animated:
ani = FuncAnimation(fig,
vis_detections_animate,
frames=xrange(len(ax.patches)),
interval=500,
blit=True)
ani.save('/tmp/animation.gif', writer='imagemagick', fps=1)
else:
plt.draw()
def nms(dets, thresh, force_cpu=False):
"""Dispatch to either CPU or GPU NMS implementations."""
if dets.shape[0] == 0:
return []
# if cfg.USE_GPU_NMS and not force_cpu:
# return gpu_nms(dets, thresh, device_id=cfg.GPU_ID)
# else:
# return cpu_nms(dets, thresh)
return py_cpu_nms(dets=dets, thresh=thresh)