From 92ed4cb2ea68be44b1ff153e00410c2082ee62df Mon Sep 17 00:00:00 2001
From: natanielruiz <nataniel777@hotmail.com>
Date: 星期二, 15 八月 2017 00:28:28 +0800
Subject: [PATCH] New experiments with hourglass
---
code/datasets.py | 222 +++++++++++++++++++++++++++++++++---------------------
1 files changed, 135 insertions(+), 87 deletions(-)
diff --git a/code/datasets.py b/code/datasets.py
index 06cd433..63ee287 100644
--- a/code/datasets.py
+++ b/code/datasets.py
@@ -7,8 +7,12 @@
import utils
+def stack_grayscale_tensor(tensor):
+ tensor = torch.cat([tensor, tensor, tensor], 0)
+ return tensor
+
class Pose_300W_LP(Dataset):
- def __init__(self, data_dir, filename_path, transform, img_ext='.jpg', annot_ext='.mat'):
+ def __init__(self, data_dir, filename_path, transform, img_ext='.jpg', annot_ext='.mat', image_mode='RGB'):
self.data_dir = data_dir
self.transform = transform
self.img_ext = img_ext
@@ -18,26 +22,55 @@
self.X_train = filename_list
self.y_train = filename_list
+ self.image_mode = image_mode
self.length = len(filename_list)
def __getitem__(self, index):
img = Image.open(os.path.join(self.data_dir, self.X_train[index] + self.img_ext))
- img = img.convert('RGB')
+ img = img.convert(self.image_mode)
+ mat_path = os.path.join(self.data_dir, self.y_train[index] + self.annot_ext)
+ shape_path = os.path.join(self.data_dir, self.y_train[index] + '_shape.npy')
- pose = utils.get_ypr_from_mat(os.path.join(self.data_dir, self.y_train[index] + self.annot_ext))
- label = torch.FloatTensor(pose)
+ # Crop the face
+ pt2d = utils.get_pt2d_from_mat(mat_path)
+ x_min = min(pt2d[0,:])
+ y_min = min(pt2d[1,:])
+ x_max = max(pt2d[0,:])
+ y_max = max(pt2d[1,:])
+
+ k = 0.15
+ x_min -= k * abs(x_max - x_min)
+ y_min -= 4 * k * abs(y_max - y_min)
+ x_max += k * abs(x_max - x_min)
+ y_max += 0.4 * k * abs(y_max - y_min)
+ img = img.crop((int(x_min), int(y_min), int(x_max), int(y_max)))
+
+ # We get the pose in radians
+ pose = utils.get_ypr_from_mat(mat_path)
+ # And convert to degrees.
+ pitch = pose[0] * 180 / np.pi
+ yaw = pose[1] * 180 / np.pi
+ roll = pose[2] * 180 / np.pi
+ # Bin values
+ bins = np.array(range(-99, 102, 3))
+ binned_pose = np.digitize([yaw, pitch, roll], bins) - 1
+
+ # Get shape
+ shape = np.load(shape_path)
+
+ labels = torch.LongTensor(np.concatenate((binned_pose, shape), axis = 0))
if self.transform is not None:
img = self.transform(img)
- return img, label, self.X_train[index]
+ return img, labels, self.X_train[index]
def __len__(self):
# 122,450
return self.length
class AFLW2000(Dataset):
- def __init__(self, data_dir, filename_path, transform, img_ext='.jpg', annot_ext='.mat'):
+ def __init__(self, data_dir, filename_path, transform, img_ext='.jpg', annot_ext='.mat', image_mode='RGB'):
self.data_dir = data_dir
self.transform = transform
self.img_ext = img_ext
@@ -47,91 +80,12 @@
self.X_train = filename_list
self.y_train = filename_list
+ self.image_mode = image_mode
self.length = len(filename_list)
def __getitem__(self, index):
img = Image.open(os.path.join(self.data_dir, self.X_train[index] + self.img_ext))
- img = img.convert('RGB')
-
- pose = utils.get_ypr_from_mat(os.path.join(self.data_dir, self.y_train[index] + self.annot_ext))
- label = torch.FloatTensor(pose)
-
- if self.transform is not None:
- img = self.transform(img)
-
- return img, label, self.X_train[index]
-
- def __len__(self):
- # 2,000
- return self.length
-
-class Pose_300W_LP_binned(Dataset):
- def __init__(self, data_dir, filename_path, transform, img_ext='.jpg', annot_ext='.mat'):
- self.data_dir = data_dir
- self.transform = transform
- self.img_ext = img_ext
- self.annot_ext = annot_ext
-
- filename_list = get_list_from_filenames(filename_path)
-
- self.X_train = filename_list
- self.y_train = filename_list
- self.length = len(filename_list)
-
- def __getitem__(self, index):
- img = Image.open(os.path.join(self.data_dir, self.X_train[index] + self.img_ext))
- img = img.convert('RGB')
- mat_path = os.path.join(self.data_dir, self.y_train[index] + self.annot_ext)
-
- # Crop the face
- pt2d = utils.get_pt2d_from_mat(mat_path)
- x_min = min(pt2d[0,:])
- y_min = min(pt2d[1,:])
- x_max = max(pt2d[0,:])
- y_max = max(pt2d[1,:])
-
- k = 0.15
- x_min -= k * abs(x_max - x_min)
- y_min -= 4 * k * abs(y_max - y_min)
- x_max += k * abs(x_max - x_min)
- y_max += 0.4 * k * abs(y_max - y_min)
- img = img.crop((int(x_min), int(y_min), int(x_max), int(y_max)))
-
- # We get the pose in radians
- pose = utils.get_ypr_from_mat(mat_path)
- # And convert to degrees.
- pitch = pose[0] * 180 / np.pi
- yaw = pose[1] * 180 / np.pi
- roll = pose[2] * 180 / np.pi
- # Bin values
- bins = np.array(range(-99, 102, 3))
- labels = torch.LongTensor(np.digitize([yaw, pitch, roll], bins) - 1)
-
- if self.transform is not None:
- img = self.transform(img)
-
- return img, labels, self.X_train[index]
-
- def __len__(self):
- # 122,450
- return self.length
-
-class AFLW2000_binned(Dataset):
- def __init__(self, data_dir, filename_path, transform, img_ext='.jpg', annot_ext='.mat'):
- self.data_dir = data_dir
- self.transform = transform
- self.img_ext = img_ext
- self.annot_ext = annot_ext
-
- filename_list = get_list_from_filenames(filename_path)
-
- self.X_train = filename_list
- self.y_train = filename_list
- self.length = len(filename_list)
-
- def __getitem__(self, index):
- img = Image.open(os.path.join(self.data_dir, self.X_train[index] + self.img_ext))
- img = img.convert('RGB')
+ img = img.convert(self.image_mode)
mat_path = os.path.join(self.data_dir, self.y_train[index] + self.annot_ext)
# Crop the face
@@ -167,6 +121,100 @@
# 2,000
return self.length
+class AFLW(Dataset):
+ def __init__(self, data_dir, filename_path, transform, img_ext='.jpg', annot_ext='.txt', image_mode='RGB'):
+ self.data_dir = data_dir
+ self.transform = transform
+ self.img_ext = img_ext
+ self.annot_ext = annot_ext
+
+ filename_list = get_list_from_filenames(filename_path)
+
+ self.X_train = filename_list
+ self.y_train = filename_list
+ self.image_mode = image_mode
+ self.length = len(filename_list)
+
+ def __getitem__(self, index):
+ img = Image.open(os.path.join(self.data_dir, self.X_train[index] + self.img_ext))
+ img = img.convert(self.image_mode)
+ txt_path = os.path.join(self.data_dir, self.y_train[index] + self.annot_ext)
+
+ # We get the pose in radians
+ annot = open(txt_path, 'r')
+ line = annot.readline().split(' ')
+ pose = [float(line[1]), float(line[2]), float(line[3])]
+ # And convert to degrees.
+ yaw = pose[0] * 180 / np.pi
+ pitch = pose[1] * 180 / np.pi
+ roll = pose[2] * 180 / np.pi
+ # Something weird with the roll in AFLW
+ if yaw < 0:
+ roll *= -1
+ # Bin values
+ bins = np.array(range(-99, 102, 3))
+ labels = torch.LongTensor(np.digitize([yaw, pitch, roll], bins) - 1)
+
+ if self.transform is not None:
+ img = self.transform(img)
+
+ return img, labels, self.X_train[index]
+
+ def __len__(self):
+ # train: 18,863
+ # test: 1,966
+ return self.length
+
+class AFW(Dataset):
+ def __init__(self, data_dir, filename_path, transform, img_ext='.jpg', annot_ext='.txt', image_mode='RGB'):
+ self.data_dir = data_dir
+ self.transform = transform
+ self.img_ext = img_ext
+ self.annot_ext = annot_ext
+
+ filename_list = get_list_from_filenames(filename_path)
+
+ self.X_train = filename_list
+ self.y_train = filename_list
+ self.image_mode = image_mode
+ self.length = len(filename_list)
+
+ def __getitem__(self, index):
+ txt_path = os.path.join(self.data_dir, self.y_train[index] + self.annot_ext)
+ img_name = self.X_train[index].split('_')[0]
+
+ img = Image.open(os.path.join(self.data_dir, img_name + self.img_ext))
+ img = img.convert(self.image_mode)
+ txt_path = os.path.join(self.data_dir, self.y_train[index] + self.annot_ext)
+
+ # We get the pose in degrees
+ annot = open(txt_path, 'r')
+ line = annot.readline().split(' ')
+ yaw, pitch, roll = [float(line[1]), float(line[2]), float(line[3])]
+
+ # Crop the face
+ margin = 40
+ x_min = float(line[4]) - margin
+ y_min = float(line[5]) - margin
+ x_max = float(line[6]) + margin
+ y_max = float(line[7]) + margin
+
+ img = img.crop((int(x_min), int(y_min), int(x_max), int(y_max)))
+
+ # Bin values
+ bins = np.array(range(-99, 102, 3))
+ labels = torch.LongTensor(np.digitize([yaw, pitch, roll], bins) - 1)
+
+ if self.transform is not None:
+ img = self.transform(img)
+
+ return img, labels, self.X_train[index]
+
+ def __len__(self):
+ # Around 200
+ return self.length
+
+
def get_list_from_filenames(file_path):
# input: relative path to .txt file with file names
# output: list of relative path names
--
Gitblit v1.8.0