def display_dataset():
logging.basicConfig(level=logging.DEBUG)
cfg = load_config()
dataset = dataset_create(cfg)
dataset.set_shuffle(False)
while True:
batch = dataset.next_batch()
for frame_id in range(1):
img = batch[Batch.inputs][frame_id, :, :, :]
img = np.squeeze(img).astype('uint8')
scmap = batch[Batch.part_score_targets][frame_id, :, :, :]
scmap = np.squeeze(scmap)
# scmask = batch[Batch.part_score_weights]
# if scmask.size > 1:
# scmask = np.squeeze(scmask).astype('uint8')
# else:
# scmask = np.zeros(img.shape)
subplot_height = 4
subplot_width = 5
num_plots = subplot_width * subplot_height
f, axarr = plt.subplots(subplot_height, subplot_width)
for j in range(num_plots):
plot_j = j // subplot_width
plot_i = j % subplot_width
curr_plot = axarr[plot_j, plot_i]
curr_plot.axis('off')
if j >= cfg.num_joints:
continue
scmap_part = scmap[:, :, j]
scmap_part = imresize(scmap_part, 8.0, interp='nearest')
scmap_part = np.lib.pad(scmap_part, ((4, 0), (4, 0)),
'minimum')
curr_plot.set_title("{}".format(j + 1))
curr_plot.imshow(img)
curr_plot.hold(True)
curr_plot.imshow(scmap_part, alpha=.5)
# figure(0)
# plt.imshow(np.sum(scmap, axis=2))
# plt.figure(100)
# plt.imshow(img)
# plt.figure(2)
# plt.imshow(scmask)
plt.show()
plt.waitforbuttonpress()
def display_dataset():
logging.basicConfig(level=logging.DEBUG)
cfg = load_config()
dataset = dataset_create(cfg)
dataset.set_shuffle(False)
while True:
batch = dataset.next_batch()
for frame_id in range(1):
img = batch[Batch.inputs][frame_id,:,:,:]
img = np.squeeze(img).astype('uint8')
scmap = batch[Batch.part_score_targets][frame_id,:,:,:]
scmap = np.squeeze(scmap)
# scmask = batch[Batch.part_score_weights]
# if scmask.size > 1:
# scmask = np.squeeze(scmask).astype('uint8')
# else:
# scmask = np.zeros(img.shape)
subplot_height = 4
subplot_width = 5
num_plots = subplot_width * subplot_height
f, axarr = plt.subplots(subplot_height, subplot_width)
for j in range(num_plots):
plot_j = j // subplot_width
plot_i = j % subplot_width
curr_plot = axarr[plot_j, plot_i]
curr_plot.axis('off')
if j >= cfg.num_joints:
continue
scmap_part = scmap[:,:,j]
scmap_part = imresize(scmap_part, 8.0, interp='nearest')
scmap_part = np.lib.pad(scmap_part, ((4, 0), (4, 0)), 'minimum')
curr_plot.set_title("{}".format(j+1))
curr_plot.imshow(img)
curr_plot.hold(True)
curr_plot.imshow(scmap_part, alpha=.5)
# figure(0)
# plt.imshow(np.sum(scmap, axis=2))
# plt.figure(100)
# plt.imshow(img)
# plt.figure(2)
# plt.imshow(scmask)
plt.show()
plt.waitforbuttonpress()
def eval_pck(cfg):
dataset = dataset_create(cfg)
filename = 'predictions.mat'
pred = sio.loadmat(filename)
joints = pred['joints']
pck_ratio_thresh = cfg.pck_threshold
num_joints = cfg.num_joints
num_images = joints.shape[1]
pred_joints = np.zeros((num_images, num_joints, 2))
gt_joints = np.zeros((num_images, num_joints, 2))
pck_thresh = np.zeros((num_images, 1))
gt_present_joints = np.zeros((num_images, num_joints))
for k in range(num_images):
pred = joints[0, k]
gt = dataset.data[k].joints[0]
if gt.shape[0] == 0:
continue
gt_joint_ids = gt[:, 0].astype('int32')
rect = enclosing_rect(gt[:, 1:3])
pck_thresh[k] = pck_ratio_thresh * np.amax(rect_size(rect))
gt_present_joints[k, gt_joint_ids] = 1
gt_joints[k, gt_joint_ids, :] = gt[:, 1:3]
pred_joints[k, :, :] = pred[:, 0:2]
dists = np.sqrt(np.sum((pred_joints - gt_joints)**2, axis=2))
correct = dists <= pck_thresh
num_all = np.sum(gt_present_joints, axis=0)
num_correct = np.zeros((num_joints, ))
for j_id in range(num_joints):
num_correct[j_id] = np.sum(correct[gt_present_joints[:, j_id] == 1,
j_id],
axis=0)
pck = num_correct / num_all * 100.0
print_results(pck, cfg)
def eval_pck(cfg):
dataset = dataset_create(cfg)
filename = 'predictions.mat'
pred = sio.loadmat(filename)
joints = pred['joints']
pck_ratio_thresh = cfg.pck_threshold
num_joints = cfg.num_joints
num_images = joints.shape[1]
pred_joints = np.zeros((num_images, num_joints, 2))
gt_joints = np.zeros((num_images, num_joints, 2))
pck_thresh = np.zeros((num_images, 1))
gt_present_joints = np.zeros((num_images, num_joints))
for k in range(num_images):
pred = joints[0,k]
gt = dataset.data[k].joints[0]
if gt.shape[0] == 0:
continue
gt_joint_ids = gt[:, 0].astype('int32')
rect = enclosing_rect(gt[:,1:3])
pck_thresh[k] = pck_ratio_thresh*np.amax(rect_size(rect))
gt_present_joints[k, gt_joint_ids] = 1
gt_joints[k, gt_joint_ids, :] = gt[:,1:3]
pred_joints[k, :, :] = pred[:,0:2]
dists = np.sqrt(np.sum((pred_joints - gt_joints)**2, axis=2))
correct = dists <= pck_thresh
num_all = np.sum(gt_present_joints, axis=0)
num_correct = np.zeros((num_joints, ))
for j_id in range(num_joints):
num_correct[j_id] = np.sum(correct[gt_present_joints[:,j_id] == 1, j_id], axis=0)
pck = num_correct/num_all*100.0
print_results(pck, cfg)
def display_dataset():
# 该文件可以显示训练之后的heatmap
# 开启日志
logging.basicConfig(level=logging.DEBUG)
# 读取配置
cfg = load_config()
# 创建数据集读取类
dataset = dataset_create(cfg)
# 不需要对数据进行洗牌
dataset.set_shuffle(False)
while True:
# 获得一批数据
batch = dataset.next_batch()
for frame_id in range(1):
# 图像
img = batch[Batch.inputs][frame_id, :, :, :]
# 转换图像类型为uint8,8位无符号整型，范围在0-255之间
# 正好为图像中像素值的区间范围
img = np.squeeze(img).astype('uint8')
# 获取heatmap
scmap = batch[Batch.part_score_targets][frame_id, :, :, :]
# 去掉多余的维度
scmap = np.squeeze(scmap)
# scmask = batch[Batch.part_score_weights]
# if scmask.size > 1:
# scmask = np.squeeze(scmask).astype('uint8')
# else:
# scmask = np.zeros(img.shape)
# 画图，这里在一个窗口中画多个图所以要定义这个窗口有几行几列
# 几行
subplot_height = 4
# 几列
subplot_width = 5
# 要画几次
num_plots = subplot_width * subplot_height
# 创建若干画图句柄
f, axarr = plt.subplots(subplot_height, subplot_width)
for j in range(num_plots):
# 需要在第几行第几列画图
plot_j = j // subplot_width
plot_i = j % subplot_width
# 获得当前的第plot_j行，第plot_i列对应的画图句柄
curr_plot = axarr[plot_j, plot_i]
# 将对应的坐标系的显示关闭
curr_plot.axis('off')
# 如果超过了需要显示的关节的个数则忽略
if j >= cfg.num_joints:
continue
# 将第j个关节的heatmap取出来
scmap_part = scmap[:, :, j]
# 缩放heatmap为8x8的，并且使用最近邻插值进行处理
scmap_part = imresize(scmap_part, 8.0, interp='nearest')
#
scmap_part = np.lib.pad(scmap_part, ((4, 0), (4, 0)),
'minimum')
# 当前画的标题是关节的编号
curr_plot.set_title("{}".format(j + 1))
# 显示图片
curr_plot.imshow(img)
# hold(True)表示在上面显示出来的图片基础之上继续画
curr_plot.hold(True)
# 将关节的heatmap也画出来，设定透明度为0.5
curr_plot.imshow(scmap_part, alpha=.5)
# figure(0)
# plt.imshow(np.sum(scmap, axis=2))
# plt.figure(100)
# plt.imshow(img)
# plt.figure(2)
# plt.imshow(scmask)
# 显示窗口
plt.show()
# 等待按键
plt.waitforbuttonpress()
