def demo_net(detector, image_name):
"""
wrapper for detector
:param detector: Detector
:param image_name: image name
:return: None
"""
# load demo data
im = cv2.imread(image_name + '.jpg')
im_array, im_scale = resize(im, config.TEST.SCALES[0], config.TRAIN.MAX_SIZE)
im_array = transform(im_array, config.PIXEL_MEANS)
roi_array = sio.loadmat(image_name + '_boxes.mat')['boxes']
batch_index_array = np.zeros((roi_array.shape[0], 1))
projected_rois = roi_array * im_scale
roi_array = np.hstack((batch_index_array, projected_rois))
scores, boxes = detector.im_detect(im_array, roi_array)
all_boxes = [[] for _ in CLASSES]
CONF_THRESH = 0.8
NMS_THRESH = 0.3
for cls in CLASSES:
cls_ind = CLASSES.index(cls)
cls_boxes = boxes[:, 4 * cls_ind:4 * (cls_ind + 1)]
cls_scores = scores[:, cls_ind]
keep = np.where(cls_scores >= CONF_THRESH)[0]
cls_boxes = cls_boxes[keep, :]
cls_scores = cls_scores[keep]
dets = np.hstack((cls_boxes, cls_scores[:, np.newaxis])).astype(np.float32)
keep = nms(dets, NMS_THRESH)
all_boxes[cls_ind] = dets[keep, :]
boxes_this_image = [[]] + [all_boxes[j] for j in range(1, len(CLASSES))]
vis_all_detection(im_array, boxes_this_image, CLASSES, 0)
def demo_net(detector, image_name):
"""
wrapper for detector
:param detector: Detector
:param image_name: image name
:return: None
"""
config.TEST.HAS_RPN = True
assert os.path.exists(image_name), image_name + ' not found'
im = cv2.imread(image_name)
im_array, im_scale = resize(im, config.SCALES[0], config.MAX_SIZE)
im_array = transform(im_array, config.PIXEL_MEANS)
im_info = np.array([[im_array.shape[2], im_array.shape[3], im_scale]], dtype=np.float32)
scores, boxes = detector.im_detect(im_array, im_info)
all_boxes = [[] for _ in CLASSES]
CONF_THRESH = 0.8
NMS_THRESH = 0.3
for cls in CLASSES:
cls_ind = CLASSES.index(cls)
cls_boxes = boxes[:, 4 * cls_ind:4 * (cls_ind + 1)]
cls_scores = scores[:, cls_ind]
keep = np.where(cls_scores >= CONF_THRESH)[0]
cls_boxes = cls_boxes[keep, :]
cls_scores = cls_scores[keep]
dets = np.hstack((cls_boxes, cls_scores[:, np.newaxis])).astype(np.float32)
keep = nms(dets.astype(np.float32), NMS_THRESH)
all_boxes[cls_ind] = dets[keep, :]
boxes_this_image = [[]] + [all_boxes[j] for j in range(1, len(CLASSES))]
vis_all_detection(im_array, boxes_this_image, CLASSES, 0)
def demo_net(detector, image_name):
"""
wrapper for detector
:param detector: Detector
:param image_name: image name
:return: None
"""
config.TEST.HAS_RPN = True
assert os.path.exists(image_name), image_name + ' not found'
im = cv2.imread(image_name)
im_array, im_scale = resize(im, config.SCALES[0], config.MAX_SIZE)
im_array = transform(im_array, config.PIXEL_MEANS)
im_info = np.array([[im_array.shape[2], im_array.shape[3], im_scale]],
dtype=np.float32)
scores, boxes = detector.im_detect(im_array, im_info)
all_boxes = [[] for _ in CLASSES]
CONF_THRESH = 0.8
NMS_THRESH = 0.3
for cls in CLASSES:
cls_ind = CLASSES.index(cls)
cls_boxes = boxes[:, 4 * cls_ind:4 * (cls_ind + 1)]
cls_scores = scores[:, cls_ind]
keep = np.where(cls_scores >= CONF_THRESH)[0]
cls_boxes = cls_boxes[keep, :]
cls_scores = cls_scores[keep]
dets = np.hstack(
(cls_boxes, cls_scores[:, np.newaxis])).astype(np.float32)
keep = nms(dets.astype(np.float32), NMS_THRESH)
all_boxes[cls_ind] = dets[keep, :]
boxes_this_image = [[]] + [all_boxes[j] for j in range(1, len(CLASSES))]
vis_all_detection(im_array, boxes_this_image, CLASSES, 0)
def main():
color = cv2.imread(args.img) # read image in b,g,r order
img, scale = resize(color.copy(), 640, 1024)
im_info = np.array([[img.shape[0], img.shape[1], scale]],
dtype=np.float32) # (h, w, scale)
img = np.swapaxes(img, 0, 2)
img = np.swapaxes(img, 1, 2) # change to r,g,b order
img = img[np.newaxis, :] # extend to (n, c, h, w)
ctx = mx.gpu(args.gpu)
_, arg_params, aux_params = mx.model.load_checkpoint(
args.prefix, args.epoch)
arg_params, aux_params = ch_dev(arg_params, aux_params, ctx)
if 'resnet' in args.prefix:
sym = resnet_50(num_class=2,
bn_mom=0.99,
bn_global=True,
is_train=False)
else:
sym = get_vgg_test(num_classes=2)
arg_params["data"] = mx.nd.array(img, ctx)
arg_params["im_info"] = mx.nd.array(im_info, ctx)
exe = sym.bind(ctx,
arg_params,
args_grad=None,
grad_req="null",
aux_states=aux_params)
exe.forward(is_train=False)
output_dict = {
name: nd
for name, nd in zip(sym.list_outputs(), exe.outputs)
}
rois = output_dict['rpn_rois_output'].asnumpy(
)[:, 1:] # first column is index
scores = output_dict['cls_prob_reshape_output'].asnumpy()[0]
bbox_deltas = output_dict['bbox_pred_reshape_output'].asnumpy()[0]
pred_boxes = bbox_pred(rois, bbox_deltas)
pred_boxes = clip_boxes(pred_boxes, (im_info[0][0], im_info[0][1]))
cls_boxes = pred_boxes[:, 4:8]
cls_scores = scores[:, 1]
keep = np.where(cls_scores >= args.thresh)[0]
cls_boxes = cls_boxes[keep, :]
cls_scores = cls_scores[keep]
dets = np.hstack((cls_boxes, cls_scores[:, np.newaxis])).astype(np.float32)
keep = nms(dets.astype(np.float32), args.nms_thresh)
dets = dets[keep, :]
keep = nest(dets, thresh=args.nest_thresh)
dets = dets[keep, :]
for i in range(dets.shape[0]):
bbox = dets[i, :4]
cv2.rectangle(
color, (int(round(bbox[0] / scale)), int(round(bbox[1] / scale))),
(int(round(bbox[2] / scale)), int(round(bbox[3] / scale))),
(0, 255, 0), 2)
cv2.imwrite("result.jpg", color)
def main():
color = cv2.imread(args.img) # read image in b,g,r order
img, scale = resize(color.copy(), 640, 1024)
im_info = np.array([[img.shape[0], img.shape[1], scale]], dtype=np.float32) # (h, w, scale)
img = np.swapaxes(img, 0, 2)
img = np.swapaxes(img, 1, 2) # change to r,g,b order
img = img[np.newaxis, :] # extend to (n, c, h, w)
ctx = mx.gpu(args.gpu)
_, arg_params, aux_params = mx.model.load_checkpoint(args.prefix, args.epoch)
arg_params, aux_params = ch_dev(arg_params, aux_params, ctx)
if 'resnet' in args.prefix:
sym = resnet_50(num_class=2, bn_mom=0.99, bn_global=True, is_train=False)
else:
sym = get_vgg_test(num_classes=2)
arg_params["data"] = mx.nd.array(img, ctx)
arg_params["im_info"] = mx.nd.array(im_info, ctx)
exe = sym.bind(ctx, arg_params, args_grad=None, grad_req="null", aux_states=aux_params)
exe.forward(is_train=False)
output_dict = {name: nd for name, nd in zip(sym.list_outputs(), exe.outputs)}
rois = output_dict['rpn_rois_output'].asnumpy()[:, 1:] # first column is index
scores = output_dict['cls_prob_reshape_output'].asnumpy()[0]
bbox_deltas = output_dict['bbox_pred_reshape_output'].asnumpy()[0]
pred_boxes = bbox_pred(rois, bbox_deltas)
pred_boxes = clip_boxes(pred_boxes, (im_info[0][0], im_info[0][1]))
cls_boxes = pred_boxes[:, 4:8]
cls_scores = scores[:, 1]
keep = np.where(cls_scores >= args.thresh)[0]
cls_boxes = cls_boxes[keep, :]
cls_scores = cls_scores[keep]
dets = np.hstack((cls_boxes, cls_scores[:, np.newaxis])).astype(np.float32)
keep = nms(dets.astype(np.float32), args.nms_thresh)
dets = dets[keep, :]
keep = nest(dets, thresh=args.nest_thresh)
dets = dets[keep, :]
for i in range(dets.shape[0]):
bbox = dets[i, :4]
cv2.rectangle(color, (int(round(bbox[0]/scale)), int(round(bbox[1]/scale))),
(int(round(bbox[2]/scale)), int(round(bbox[3]/scale))), (0, 255, 0), 2)
cv2.imwrite("result.jpg", color)
def demo_net(detector, image_name):
"""
wrapper for detector
:param detector: Detector
:param image_name: image name
:return: None
"""
# load demo data
im = cv2.imread(image_name + '.jpg')
im_array, im_scale = resize(im, config.TEST.SCALES[0],
config.TRAIN.MAX_SIZE)
im_array = transform(im_array, config.PIXEL_MEANS)
roi_array = sio.loadmat(image_name + '_boxes.mat')['boxes']
batch_index_array = np.zeros((roi_array.shape[0], 1))
projected_rois = roi_array * im_scale
roi_array = np.hstack((batch_index_array, projected_rois))
scores, boxes = detector.im_detect(im_array, roi_array)
all_boxes = [[] for _ in CLASSES]
CONF_THRESH = 0.8
NMS_THRESH = 0.3
for cls in CLASSES:
cls_ind = CLASSES.index(cls)
cls_boxes = boxes[:, 4 * cls_ind:4 * (cls_ind + 1)]
cls_scores = scores[:, cls_ind]
keep = np.where(cls_scores >= CONF_THRESH)[0]
cls_boxes = cls_boxes[keep, :]
cls_scores = cls_scores[keep]
dets = np.hstack(
(cls_boxes, cls_scores[:, np.newaxis])).astype(np.float32)
keep = nms(dets, NMS_THRESH)
all_boxes[cls_ind] = dets[keep, :]
boxes_this_image = [[]] + [all_boxes[j] for j in range(1, len(CLASSES))]
vis_all_detection(im_array, boxes_this_image, CLASSES, 0)
def pred_eval(detector, test_data, imdb, vis=False):
"""
wrapper for calculating offline validation for faster data analysis
in this example, all threshold are set by hand
:param detector: Detector
:param test_data: data iterator, must be non-shuffle
:param imdb: image database
:param vis: controls visualization
:return:
"""
assert not test_data.shuffle
thresh = 0.1
# limit detections to max_per_image over all classes
max_per_image = 100
num_images = imdb.num_images
# all detections are collected into:
# all_boxes[cls][image] = N x 5 array of detections in
# (x1, y1, x2, y2, score)
all_boxes = [[[] for _ in xrange(num_images)]
for _ in xrange(imdb.num_classes)]
i = 0
for databatch in test_data:
if i % 10 == 0:
print 'testing {}/{}'.format(i, imdb.num_images)
scores, boxes = detector.im_detect(databatch.data['data'], databatch.data['rois'])
# we used scaled image & roi to train, so it is necessary to transform them back
# visualization should also be from the original size
im_path = imdb.image_path_from_index(imdb.image_set_index[i])
im = cv2.imread(im_path)
im_height = im.shape[0]
scale = float(databatch.data['data'].shape[2]) / float(im_height)
im = image_processing.transform(im, config.PIXEL_MEANS)
for j in range(1, imdb.num_classes):
indexes = np.where(scores[:, j] > thresh)[0]
cls_scores = scores[indexes, j]
cls_boxes = boxes[indexes, j * 4:(j + 1) * 4] / scale
cls_dets = np.hstack((cls_boxes, cls_scores[:, np.newaxis]))
keep = nms(cls_dets, config.TEST.NMS)
all_boxes[j][i] = cls_dets[keep, :]
if max_per_image > 0:
image_scores = np.hstack([all_boxes[j][i][:, -1]
for j in range(1, imdb.num_classes)])
if len(image_scores) > max_per_image:
image_thresh = np.sort(image_scores)[-max_per_image]
for j in range(1, imdb.num_classes):
keep = np.where(all_boxes[j][i][:, -1] >= image_thresh)[0]
all_boxes[j][i] = all_boxes[j][i][keep, :]
boxes_this_image = [[]] + [all_boxes[j][i] for j in range(1, imdb.num_classes)]
if vis:
vis_all_detection(im, boxes_this_image,
imdb_classes=imdb.classes)
i += 1
cache_folder = os.path.join(imdb.cache_path, imdb.name)
if not os.path.exists(cache_folder):
os.mkdir(cache_folder)
det_file = os.path.join(cache_folder, 'detections.pkl')
with open(det_file, 'wb') as f:
cPickle.dump(all_boxes, f)
imdb.evaluate_detections(all_boxes)
def forward(self, is_train, req, in_data, out_data, aux):
# for each (H, W) location i
# generate A anchor boxes centered on cell i
# apply predicted bbox deltas at cell i to each of the A anchors
# clip predicted boxes to image
# remove predicted boxes with either height or width < threshold
# sort all (proposal, score) pairs by score from highest to lowest
# take top pre_nms_topN proposals before NMS
# apply NMS with threshold 0.7 to remaining proposals
# take after_nms_topN proposals after NMS
# return the top proposals (-> RoIs top, scores top)
pre_nms_topN = config[self.cfg_key].RPN_PRE_NMS_TOP_N
post_nms_topN = config[self.cfg_key].RPN_POST_NMS_TOP_N
nms_thresh = config[self.cfg_key].RPN_NMS_THRESH
min_size = config[self.cfg_key].RPN_MIN_SIZE
# the first set of anchors are background probabilities
# keep the second part
scores = in_data[0].asnumpy()[:, self._num_anchors:, :, :]
bbox_deltas = in_data[1].asnumpy()
im_info = in_data[2].asnumpy()[0, :]
if DEBUG:
print 'im_size: ({}, {})'.format(im_info[0], im_info[1])
print 'scale: {}'.format(im_info[2])
# 1. Generate proposals from bbox_deltas and shifted anchors
height, width = scores.shape[-2:]
if DEBUG:
print 'score map size: {}'.format(scores.shape)
# Enumerate all shifts
shift_x = np.arange(0, width) * self._feat_stride
shift_y = np.arange(0, height) * self._feat_stride
shift_x, shift_y = np.meshgrid(shift_x, shift_y)
shifts = np.vstack((shift_x.ravel(), shift_y.ravel(), shift_x.ravel(),
shift_y.ravel())).transpose()
# Enumerate all shifted anchors:
#
# add A anchors (1, A, 4) to
# cell K shifts (K, 1, 4) to get
# shift anchors (K, A, 4)
# reshape to (K*A, 4) shifted anchors
A = self._num_anchors
K = shifts.shape[0]
anchors = self._anchors.reshape((1, A, 4)) + shifts.reshape(
(1, K, 4)).transpose((1, 0, 2))
anchors = anchors.reshape((K * A, 4))
# Transpose and reshape predicted bbox transformations to get them
# into the same order as the anchors:
#
# bbox deltas will be (1, 4 * A, H, W) format
# transpose to (1, H, W, 4 * A)
# reshape to (1 * H * W * A, 4) where rows are ordered by (h, w, a)
# in slowest to fastest order
bbox_deltas = bbox_deltas.transpose((0, 2, 3, 1)).reshape((-1, 4))
# Same story for the scores:
#
# scores are (1, A, H, W) format
# transpose to (1, H, W, A)
# reshape to (1 * H * W * A, 1) where rows are ordered by (h, w, a)
scores = scores.transpose((0, 2, 3, 1)).reshape((-1, 1))
# Convert anchors into proposals via bbox transformations
proposals = bbox_pred(anchors, bbox_deltas)
# 2. clip predicted boxes to image
proposals = clip_boxes(proposals, im_info[:2])
# 3. remove predicted boxes with either height or width < threshold
# (NOTE: convert min_size to input image scale stored in im_info[2])
keep = ProposalOperator._filter_boxes(proposals, min_size * im_info[2])
proposals = proposals[keep, :]
scores = scores[keep]
# 4. sort all (proposal, score) pairs by score from highest to lowest
# 5. take top pre_nms_topN (e.g. 6000)
order = scores.ravel().argsort()[::-1]
if pre_nms_topN > 0:
order = order[:pre_nms_topN]
proposals = proposals[order, :]
scores = scores[order]
# 6. apply nms (e.g. threshold = 0.7)
# 7. take after_nms_topN (e.g. 300)
# 8. return the top proposals (-> RoIs top)
keep = nms(np.hstack((proposals, scores)), nms_thresh)
if post_nms_topN > 0:
keep = keep[:post_nms_topN]
# pad to ensure output size remains unchanged
if len(keep) < post_nms_topN:
pad = npr.choice(keep, size=post_nms_topN - len(keep))
keep = np.hstack((keep, pad))
proposals = proposals[keep, :]
scores = scores[keep]
# Output rois array
# Our RPN implementation only supports a single input image, so all
# batch inds are 0
batch_inds = np.zeros((proposals.shape[0], 1), dtype=np.float32)
blob = np.hstack((batch_inds, proposals.astype(np.float32,
copy=False)))
self.assign(out_data[0], req[0], blob)
if self._output_score:
self.assign(out_data[1], req[1],
scores.astype(np.float32, copy=False))
def forward(self, is_train, req, in_data, out_data, aux):
# for each (H, W) location i
# generate A anchor boxes centered on cell i
# apply predicted bbox deltas at cell i to each of the A anchors
# clip predicted boxes to image
# remove predicted boxes with either height or width < threshold
# sort all (proposal, score) pairs by score from highest to lowest
# take top pre_nms_topN proposals before NMS
# apply NMS with threshold 0.7 to remaining proposals
# take after_nms_topN proposals after NMS
# return the top proposals (-> RoIs top, scores top)
pre_nms_topN = config[self.cfg_key].RPN_PRE_NMS_TOP_N
post_nms_topN = config[self.cfg_key].RPN_POST_NMS_TOP_N
nms_thresh = config[self.cfg_key].RPN_NMS_THRESH
min_size = config[self.cfg_key].RPN_MIN_SIZE
# the first set of anchors are background probabilities
# keep the second part
scores = in_data[0].asnumpy()[:, self._num_anchors:, :, :]
if np.isnan(scores).any():
raise ValueError("there is nan in input scores")
bbox_deltas = in_data[1].asnumpy()
if np.isnan(bbox_deltas).any():
raise ValueError("there is nan in input bbox_deltas")
im_info = in_data[2].asnumpy()[0, :]
if DEBUG:
print 'im_size: ({}, {})'.format(im_info[0], im_info[1])
print 'scale: {}'.format(im_info[2])
# 1. Generate proposals from bbox_deltas and shifted anchors
height, width = scores.shape[-2:]
if self.cfg_key == 'TRAIN':
height, width = int(im_info[0] / self._feat_stride), int(im_info[1] / self._feat_stride)
if DEBUG:
print 'score map size: {}'.format(scores.shape)
print "resudial = ", scores.shape[2] - height, scores.shape[3] - width
# Enumerate all shifts
shift_x = np.arange(0, width) * self._feat_stride
shift_y = np.arange(0, height) * self._feat_stride
shift_x, shift_y = np.meshgrid(shift_x, shift_y)
shifts = np.vstack((shift_x.ravel(), shift_y.ravel(), shift_x.ravel(), shift_y.ravel())).transpose()
# Enumerate all shifted anchors:
#
# add A anchors (1, A, 4) to
# cell K shifts (K, 1, 4) to get
# shift anchors (K, A, 4)
# reshape to (K*A, 4) shifted anchors
A = self._num_anchors
K = shifts.shape[0]
anchors = self._anchors.reshape((1, A, 4)) + shifts.reshape((1, K, 4)).transpose((1, 0, 2))
anchors = anchors.reshape((K * A, 4))
# Transpose and reshape predicted bbox transformations to get them
# into the same order as the anchors:
#
# bbox deltas will be (1, 4 * A, H, W) format
# transpose to (1, H, W, 4 * A)
# reshape to (1 * H * W * A, 4) where rows are ordered by (h, w, a)
# in slowest to fastest order
bbox_deltas = clip_pad(bbox_deltas, (height, width))
bbox_deltas = bbox_deltas.transpose((0, 2, 3, 1)).reshape((-1, 4))
# Same story for the scores:
#
# scores are (1, A, H, W) format
# transpose to (1, H, W, A)
# reshape to (1 * H * W * A, 1) where rows are ordered by (h, w, a)
scores = scores.transpose((0, 2, 3, 1)).reshape((-1, 1))
# Convert anchors into proposals via bbox transformations
proposals = bbox_pred(anchors, bbox_deltas)
# 2. clip predicted boxes to image
proposals = clip_boxes(proposals, im_info[:2])
# 3. remove predicted boxes with either height or width < threshold
# (NOTE: convert min_size to input image scale stored in im_info[2])
keep = ProposalOperator._filter_boxes(proposals, min_size * im_info[2])
proposals = proposals[keep, :]
scores = scores[keep]
# 4. sort all (proposal, score) pairs by score from highest to lowest
# 5. take top pre_nms_topN (e.g. 6000)
order = scores.ravel().argsort()[::-1]
if pre_nms_topN > 0:
order = order[:pre_nms_topN]
proposals = proposals[order, :]
scores = scores[order]
# 6. apply nms (e.g. threshold = 0.7)
# 7. take after_nms_topN (e.g. 300)
# 8. return the top proposals (-> RoIs top)
keep = nms(np.hstack((proposals, scores)), nms_thresh)
if post_nms_topN > 0:
keep = keep[:post_nms_topN]
# pad to ensure output size remains unchanged
if len(keep) < post_nms_topN:
if len(keep) == 0:
logging.log(logging.ERROR, "currently len(keep) is zero")
pad = npr.choice(keep, size=post_nms_topN - len(keep))
keep = np.hstack((keep, pad))
proposals = proposals[keep, :]
scores = scores[keep]
# Output rois array
# Our RPN implementation only supports a single input image, so all
# batch inds are 0
batch_inds = np.zeros((proposals.shape[0], 1), dtype=np.float32)
blob = np.hstack((batch_inds, proposals.astype(np.float32, copy=False)))
self.assign(out_data[0], req[0], blob)
if self._output_score:
self.assign(out_data[1], req[1], scores.astype(np.float32, copy=False))
def pred_eval(detector, test_data, imdb, vis=False):
"""
wrapper for calculating offline validation for faster data analysis
in this example, all threshold are set by hand
:param detector: Detector
:param test_data: data iterator, must be non-shuffle
:param imdb: image database
:param vis: controls visualization
:return:
"""
assert not test_data.shuffle
thresh = 0.05
# limit detections to max_per_image over all classes
max_per_image = 100
num_images = imdb.num_images
# all detections are collected into:
# all_boxes[cls][image] = N x 5 array of detections in
# (x1, y1, x2, y2, score)
all_boxes = [[[] for _ in xrange(num_images)]
for _ in xrange(imdb.num_classes)]
i = 0
for databatch in test_data:
if i % 10 == 0:
print 'testing {}/{}'.format(i, imdb.num_images)
if config.TEST.HAS_RPN:
scores, boxes = detector.im_detect(databatch.data['data'], im_info=databatch.data['im_info'])
scale = databatch.data['im_info'][0, 2]
else:
scores, boxes = detector.im_detect(databatch.data['data'], roi_array=databatch.data['rois'])
# we used scaled image & roi to train, so it is necessary to transform them back
# visualization should also be from the original size
im_path = imdb.image_path_from_index(imdb.image_set_index[i])
im = cv2.imread(im_path)
im_height = im.shape[0]
scale = float(databatch.data['data'].shape[2]) / float(im_height)
for j in range(1, imdb.num_classes):
indexes = np.where(scores[:, j] > thresh)[0]
cls_scores = scores[indexes, j]
cls_boxes = boxes[indexes, j * 4:(j + 1) * 4] / scale
cls_dets = np.hstack((cls_boxes, cls_scores[:, np.newaxis]))
keep = nms(cls_dets, config.TEST.NMS)
all_boxes[j][i] = cls_dets[keep, :]
if max_per_image > 0:
image_scores = np.hstack([all_boxes[j][i][:, -1]
for j in range(1, imdb.num_classes)])
if len(image_scores) > max_per_image:
image_thresh = np.sort(image_scores)[-max_per_image]
for j in range(1, imdb.num_classes):
keep = np.where(all_boxes[j][i][:, -1] >= image_thresh)[0]
all_boxes[j][i] = all_boxes[j][i][keep, :]
boxes_this_image = [[]] + [all_boxes[j][i] for j in range(1, imdb.num_classes)]
if vis:
# visualize the testing scale
for box in boxes_this_image:
if isinstance(box, np.ndarray):
box[:, :4] *= scale
vis_all_detection(databatch.data['data'], boxes_this_image,
imdb_classes=imdb.classes)
i += 1
cache_folder = os.path.join(imdb.cache_path, imdb.name)
if not os.path.exists(cache_folder):
os.mkdir(cache_folder)
det_file = os.path.join(cache_folder, 'detections.pkl')
with open(det_file, 'wb') as f:
cPickle.dump(all_boxes, f)
imdb.evaluate_detections(all_boxes)
