def process_image_detailed(self, img_path):
img = cv2.imread(img_path)
_, cntk_img_input, dims = resize_and_pad(img, self._img_shape[2], self._img_shape[1])
# compute ROI proposals and apply scaling and padding to them
# [target_w, target_h, img_width, img_height, top, bottom, left, right, scale_factor]
img_stats = compute_image_stats(len(img[0]), len(img), self._img_shape[2], self._img_shape[1])
scale_factor = img_stats[-1]
top = img_stats[4]
left = img_stats[6]
num_proposals = self._cfg['NUM_ROI_PROPOSALS']
proposals = compute_proposals(img, num_proposals, self._cfg)
proposals = proposals * scale_factor
proposals += (left, top, left, top)
output = self._eval_model.eval({self._eval_model.arguments[0]: [cntk_img_input],
self._eval_model.arguments[1]: np.array(proposals, dtype=np.float32)})
out_dict = dict([(k.name, k) for k in output])
out_cls_pred = output[out_dict['cls_pred']][0]
out_rpn_rois = proposals
out_bbox_regr = output[out_dict['bbox_regr']][0]
return out_cls_pred, out_rpn_rois, out_bbox_regr, dims
def process_image_detailed(self, img_path):
img = cv2.imread(img_path)
_, cntk_img_input, dims = resize_and_pad(img, self._img_shape[2],
self._img_shape[1])
# compute ROI proposals and apply scaling and padding to them
# [target_w, target_h, img_width, img_height, top, bottom, left, right, scale_factor]
img_stats = compute_image_stats(len(img[0]), len(img),
self._img_shape[2], self._img_shape[1])
scale_factor = img_stats[-1]
top = img_stats[4]
left = img_stats[6]
num_proposals = self._cfg['NUM_ROI_PROPOSALS']
proposals = compute_proposals(img, num_proposals, self._cfg)
proposals = proposals * scale_factor
proposals += (left, top, left, top)
output = self._eval_model.eval({
self._eval_model.arguments[0]: [cntk_img_input],
self._eval_model.arguments[1]:
np.array(proposals, dtype=np.float32)
})
out_dict = dict([(k.name, k) for k in output])
out_cls_pred = output[out_dict['cls_pred']][0]
out_rpn_rois = proposals
out_bbox_regr = output[out_dict['bbox_regr']][0]
return out_cls_pred, out_rpn_rois, out_bbox_regr, dims
def _prepare_annotations_proposals_and_stats(self, index, img):
img_width = len(img[0])
img_height = len(img)
# prepare image statistics for scaling and padding images later
# [target_w, target_h, img_width, img_height, top, bottom, left, right, scale_factor]
img_stats = compute_image_stats(img_width, img_height, self._pad_width,
self._pad_height)
self._img_stats[index] = img_stats
scale_factor = img_stats[-1]
top = img_stats[4]
left = img_stats[6]
# prepare annotations
annotations = self._gt_annotations[index]
xyxy = annotations[:, :4]
xyxy *= scale_factor
xyxy += (left, top, left, top)
# not needed since xyxy is just a reference: annotations[:, :4] = xyxy
# TODO: do we need to round/floor/ceil xyxy coords?
annotations[:, 0] = np.round(annotations[:, 0])
annotations[:, 1] = np.round(annotations[:, 1])
annotations[:, 2] = np.round(annotations[:, 2])
annotations[:, 3] = np.round(annotations[:, 3])
# prepare proposals
if self._proposal_provider is not None:
proposals = self._proposal_provider.get_proposals(index, img)
if self._proposal_provider.requires_scaling():
proposals = proposals * scale_factor
proposals += (left, top, left, top)
self._proposal_dict[index] = proposals
if self._provide_targets:
# add gt rois to front of list for Fast R-CNN
gt_rois = annotations[np.where(annotations[:, 4] > 0)]
num_proposals = proposals.shape[0]
num_gt = gt_rois.shape[0]
proposals_incl_gt = np.zeros(proposals.shape)
proposals_incl_gt[:num_gt, :] = gt_rois[:, :4]
proposals_incl_gt[num_gt:, :] = proposals[:(num_proposals -
num_gt), :]
self._proposal_dict[index] = proposals_incl_gt
# prepare proposal targets
self._proposal_targets[index] = \
compute_targets(proposals_incl_gt, gt_rois, iou_threshold=self._proposal_iou_threshold,
normalize_means=self._normalize_means, normalize_stds=self._normalize_stds)
def _prepare_annotations_proposals_and_stats(self, index, img):
img_width = len(img[0])
img_height = len(img)
# prepare image statistics for scaling and padding images later
# [target_w, target_h, img_width, img_height, top, bottom, left, right, scale_factor]
img_stats = compute_image_stats(img_width, img_height, self._pad_width, self._pad_height)
self._img_stats[index] = img_stats
scale_factor = img_stats[-1]
top = img_stats[4]
left = img_stats[6]
# prepare annotations
annotations = self._gt_annotations[index]
xyxy = annotations[:, :4]
xyxy *= scale_factor
xyxy += (left, top, left, top)
# not needed since xyxy is just a reference: annotations[:, :4] = xyxy
# TODO: do we need to round/floor/ceil xyxy coords?
annotations[:, 0] = np.round(annotations[:, 0])
annotations[:, 1] = np.round(annotations[:, 1])
annotations[:, 2] = np.round(annotations[:, 2])
annotations[:, 3] = np.round(annotations[:, 3])
# prepare proposals
if self._proposal_provider is not None:
proposals = self._proposal_provider.get_proposals(index, img)
if self._proposal_provider.requires_scaling():
proposals = proposals * scale_factor
proposals += (left, top, left, top)
self._proposal_dict[index] = proposals
if self._provide_targets:
# add gt rois to front of list for Fast R-CNN
gt_rois = annotations[np.where(annotations[:, 4] > 0)]
num_proposals = proposals.shape[0]
num_gt = gt_rois.shape[0]
proposals_incl_gt = np.zeros(proposals.shape)
proposals_incl_gt[:num_gt,:] = gt_rois[:,:4]
proposals_incl_gt[num_gt:,:] = proposals[:(num_proposals - num_gt),:]
self._proposal_dict[index] = proposals_incl_gt
# prepare proposal targets
self._proposal_targets[index] = \
compute_targets(proposals_incl_gt, gt_rois, iou_threshold=self._proposal_iou_threshold,
normalize_means=self._normalize_means, normalize_stds=self._normalize_stds)