def im_detect(sess, net, im, timers, im_idx=None, nbr_gts=None):
# Setup image blob
blobs = {}
blobs['data'], im_scales, blobs['im_shape_orig'] = get_image_blob(im)
im_blob = blobs['data']
blobs['im_info'] = np.array([im_blob.shape[1], im_blob.shape[2], im_scales[0]])
# Run drl-RPN
scores, pred_bboxes, timers, stats \
= run_drl_rpn(sess, net, blobs, timers, 'test', cfg.DRL_RPN_TEST.BETA,
im_idx, nbr_gts)
return scores, pred_bboxes, timers, stats
def im_detect(sess, net, im, timers, im_idx=None, nbr_gts=None):
# Setup image blob
blobs = {}
blobs['data'], im_scales, blobs['im_shape_orig'] = get_image_blob(im)
im_blob = blobs['data']
blobs['im_info'] = np.array(
[im_blob.shape[1], im_blob.shape[2], im_scales[0]])
# Run LRP-HAI
scores, pred_bboxes, timers, stats \
= run_LRP_HAI(sess, net, blobs, timers, 'test', cfg.LRP_HAI_TEST.BETA,
im_idx, nbr_gts, alpha=cfg.LRP_HAI.ALPHA)
return scores, pred_bboxes, timers, stats
def visualize(im, im_idx, boxes, show_text=True):
# Setup image blob
blob = {}
blob['data'], im_scales, blob['im_shape_orig'] = get_image_blob(im)
im_blob = blob['data']
im_shape = blob['im_shape_orig']
blob['im_info'] = np.array(
[im_blob.shape[1], im_blob.shape[2], im_scales[0]])
im_info = blob['im_info']
# Make sure image in right range
im = im_blob[0, :, :, :]
im -= np.min(im)
im /= np.max(im)
im = resize(im, (im_shape[0], im_shape[1]), order=1, mode='reflect')
# BGR --> RGB
im = im[..., ::-1]
# Produce final detections post-NMS
cls_dets, names_and_coords = produce_trusted_boxes(boxes, thresh=0.70)
# Show image
fig, ax = plt.subplots(1)
ax.imshow(im)
# Draw all detection boxes
for j in range(len(names_and_coords)):
coords = names_and_coords[j]['coords']
score = names_and_coords[j]['score']
name = names_and_coords[j]['class_name']
color = names_and_coords[j]['color']
cls_det = cls_dets[j]
# Show object category + confidence
if show_text:
ax.text(coords[0],
coords[1],
name + " " + str(score),
weight='bold',
color='black',
fontsize=8,
horizontalalignment='center',
verticalalignment='center',
bbox=dict(facecolor='white', edgecolor='white', pad=-0.1))
# Show detection bounding boxes
rect = patches.Rectangle((cls_det[0], cls_det[1]),
cls_det[2] - cls_det[0],
cls_det[3] - cls_det[1],
linewidth=7,
edgecolor=color,
facecolor='none')
ax.add_patch(rect)
# Final save / close of figure
im_name = 'im' + str(im_idx + 1) + '.jpg'
plt.savefig('img-out-deepSORT/' + im_name)
plt.close()
# Display success message
print("Saved image " + im_name + "!\n")