def vis_all_joints(img: PIL, output, figsize=(8, 8), save_filename=None):
fig = plt.figure(figsize=figsize)
H, W = output[0].shape
plt.imshow(img.resize((W, H)), cmap='gray', interpolation='bicubic')
joints = np.clip(output, 0.4, 1.0)
joint_sum = np.sum(output, axis=0) # stacking all layers
plt.imshow(joint_sum, alpha=0.5, cmap='jet', interpolation='bicubic')
plt.axis('off')
if save_filename:
fig.savefig(save_filename)
return
def _pre_process_image(self, img: PIL):
img = img.resize((self.run_parameters.get('input_shape')[0],
self.run_parameters.get('input_shape')[1]),
Image.ANTIALIAS)
if self.run_parameters.get('greyscale'):
img = img.convert.convert('LA')
else:
img = img.convert('RGB')
img = np.array(img)
img = img / 255
# casts list to tuple
img_shape = (1, ) + tuple(self.run_parameters.get('input_shape'))
img = img.reshape(img_shape)
return img
def vis_joints(img: PIL, output, save_filename=None):
fig = plt.figure(figsize=(16, 16))
for idx, joint_data in enumerate(output):
ax = fig.add_subplot(4, 4, idx + 1, xticks=[], yticks=[])
plt.title(f'{idx}: {JOINT_NAMES[idx]}')
H, W = joint_data.shape
plt.imshow(img.resize((W, H)), cmap='gray', interpolation='bicubic')
plt.imshow(
joint_data,
alpha=0.5,
cmap='jet',
)
plt.axis('off')
if save_filename:
fig.savefig(save_filename)
return
def transform_image(img: PIL):
img = img.resize((256, 256), PIL.Image.BICUBIC) # resizing image
img = np.asarray(img)
# converting from HWC to CHW format
img_chw = HWC_2_CHW(img)
# Convert image to floating point in the range 0 to 1
img_chw = np.float32(img_chw) / 255.0
# Normalizing image data
means, stds = [0.485, 0.456, 0.406], [0.229, 0.224, 0.225]
img_chw = img_normalize(img_chw, means, stds)
img_chw = np.expand_dims(img_chw, axis=0) # Making batch size of 1
return img_chw
