def plot_results(self, results=None, classes=None, show_labels=True, gt_data=None, confidence_threshold=None):
if results is None:
results = self.results
if confidence_threshold is not None:
mask = results[:, 4] > confidence_threshold
results = results[mask]
if classes is not None:
colors = plt.cm.hsv(np.linspace(0, 1, len(classes)+1)).tolist()
ax = plt.gca()
im = plt.gci()
image_size = im.get_size()
# draw ground truth
if gt_data is not None:
for box in gt_data:
label = np.nonzero(box[4:])[0][0]+1
color = 'g' if classes == None else colors[label]
xy_rec = to_rec(box[:4], image_size)
ax.add_patch(plt.Polygon(xy_rec, fill=True, color=color, linewidth=1, alpha=0.3))
# draw prediction
for r in results:
label = int(r[5])
confidence = r[4]
color = 'r' if classes == None else colors[label]
xy_rec = to_rec(r[:4], image_size)
ax.add_patch(plt.Polygon(xy_rec, fill=False, edgecolor=color, linewidth=2))
if show_labels:
label_name = label if classes == None else classes[label]
xmin, ymin = xy_rec[0]
display_txt = '%0.2f, %s' % (confidence, label_name)
ax.text(xmin, ymin, display_txt, bbox={'facecolor':color, 'alpha':0.5})
def plot_assignment(self, map_idx):
ax = plt.gca()
im = plt.gci()
img_h, img_w = image_size = im.get_size()
# ground truth
boxes = self.gt_boxes
boxes_x = (boxes[:, 0] + boxes[:, 2]) / 2. * img_h
boxes_y = (boxes[:, 1] + boxes[:, 3]) / 2. * img_w
for box in boxes:
xy_rec = to_rec(box[:4], image_size)
ax.add_patch(
plt.Polygon(xy_rec, fill=False, edgecolor='b', linewidth=2))
plt.plot(boxes_x, boxes_y, 'bo', markersize=6)
# prior boxes
for idx, box_idx in self.match_indices.items():
if idx >= self.map_offsets[map_idx] and idx < self.map_offsets[
map_idx + 1]:
x, y = self.priors_xy[idx]
w, h = self.priors_wh[idx]
plt.plot(x, y, 'ro', markersize=4)
plt.plot([x, boxes_x[box_idx]], [y, boxes_y[box_idx]],
'-r',
linewidth=1)
ax.add_patch(
plt.Rectangle((x - w / 2, y - h / 2),
w + 1,
h + 1,
fill=False,
edgecolor='y',
linewidth=2))