-
Notifications
You must be signed in to change notification settings - Fork 0
/
sample_inference.py
311 lines (242 loc) · 13.8 KB
/
sample_inference.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
import cv2
import numpy as np
import open3d as o3d
from PIL import Image
from inference import Inferencer
import matplotlib.pyplot as plt
import os
from utils.data_preparation import process_data, exr_loader
DILATION_KERNEL = np.array([[0, 1, 1, 0], [1, 1, 1, 1], [1, 1, 1, 1], [0, 1, 1, 0]]).astype(np.uint8)
def dropout_random_ellipses_4corruptmask(mask, noise_params):
""" Randomly drop a few ellipses in the image for robustness.
This is adapted from the DexNet 2.0 codebase.
Their code: https://github.com/BerkeleyAutomation/gqcnn/blob/75040b552f6f7fb264c27d427b404756729b5e88/gqcnn/sgd_optimizer.py
@param depth_img: a [H x W] set of depth z values
"""
dropout_mask = mask.copy()
# Sample number of ellipses to dropout
num_ellipses_to_dropout = np.random.poisson(noise_params['ellipse_dropout_mean'])
# Sample ellipse centers
zero_pixel_indices = np.array(np.where(dropout_mask == 0)).T # Shape: [#nonzero_pixels x 2]
dropout_centers_indices = np.random.choice(zero_pixel_indices.shape[0], size=num_ellipses_to_dropout)
dropout_centers = zero_pixel_indices[dropout_centers_indices, :] # Shape: [num_ellipses_to_dropout x 2]
# Sample ellipse radii and angles
x_radii = np.random.gamma(noise_params['ellipse_gamma_shape'], noise_params['ellipse_gamma_scale'], size=num_ellipses_to_dropout)
y_radii = np.random.gamma(noise_params['ellipse_gamma_shape'], noise_params['ellipse_gamma_scale'], size=num_ellipses_to_dropout)
angles = np.random.randint(0, 360, size=num_ellipses_to_dropout)
# Dropout ellipses
for i in range(num_ellipses_to_dropout):
center = dropout_centers[i, :]
x_radius = np.round(x_radii[i]).astype(int)
y_radius = np.round(y_radii[i]).astype(int)
angle = angles[i]
# get ellipse mask
tmp_mask = np.zeros_like(dropout_mask, dtype=np.uint8)
tmp_mask = cv2.ellipse(tmp_mask, tuple(center[::-1]), (x_radius, y_radius), angle=angle, startAngle=0, endAngle=360, color=1, thickness=-1)
# update depth and corrupt mask
dropout_mask[tmp_mask == 1] = 1
return dropout_mask
def handle_depth(depth, depth_gt, depth_gt_mask):
depth[depth_gt_mask==1] = 0
depth_gt_mask_uint8 = np.where(depth < 0.000000001, 255, 0).astype(np.uint8)
depth_gt_mask_uint8[depth_gt_mask_uint8 != 0] = 1
depth_uint8 = depth.copy() / depth.max() * 255
depth_uint8 = np.array(depth_uint8, dtype=np.uint8)
depth_uint8 = cv2.inpaint(depth_uint8, depth_gt_mask_uint8, 5, cv2.INPAINT_NS)
depth_uint8 = np.array(depth_uint8, dtype=np.float32) / 255 * depth.max()
# depth_noise = np.random.randn(*(depth.shape))
# depth_noise = depth_noise / np.abs(depth_noise).max() * depth.max() / 1000.0
# depth[depth_gt_mask==1] = depth_uint8[depth_gt_mask == 1] + depth_noise[depth_gt_mask == 1]
mask_pixel_indices1 = np.array(np.where(depth_gt_mask == 1)).T # Shape: [#nonzero_pixels x 2]
dropout_size = int(mask_pixel_indices1.shape[0] * 0.003)
dropout_centers_indices = np.random.choice(mask_pixel_indices1.shape[0], size=dropout_size)
dropout_centers = mask_pixel_indices1[dropout_centers_indices, :]
x_radii = np.random.gamma(3.0, 2.0, size=dropout_size)
y_radii = np.random.gamma(3.0, 2.0, size=dropout_size)
angles = np.random.randint(0, 360, size=dropout_size)
result_mask = np.zeros_like(depth_gt_mask, dtype=np.uint8)
for i in range(dropout_size // 2):
center = dropout_centers[i, :]
x_radius = np.round(x_radii[i]).astype(int)
y_radius = np.round(y_radii[i]).astype(int)
angle = angles[i]
# get ellipse mask
tmp_mask = np.zeros_like(depth_gt_mask, dtype=np.uint8)
tmp_mask = cv2.ellipse(tmp_mask, tuple(center[::-1]), (x_radius, y_radius), angle=angle, startAngle=0, endAngle=360, color=1, thickness=-1)
# update depth and corrupt mask
result_mask[tmp_mask == 1] = 1
mask = np.logical_and(result_mask, depth_gt_mask_uint8)
depth[mask==1] = depth_uint8[mask == 1]
result_mask = np.zeros_like(depth_gt_mask, dtype=np.uint8)
for i in range(dropout_size - dropout_size // 2):
center = dropout_centers[i + dropout_size // 2, :]
x_radius = np.round(x_radii[i + dropout_size // 2]).astype(int)
y_radius = np.round(y_radii[i + dropout_size // 2]).astype(int)
angle = angles[i + dropout_size // 2]
# get ellipse mask
tmp_mask = np.zeros_like(depth_gt_mask, dtype=np.uint8)
tmp_mask = cv2.ellipse(tmp_mask, tuple(center[::-1]), (x_radius, y_radius), angle=angle, startAngle=0, endAngle=360, color=1, thickness=-1)
# update depth and corrupt mask
result_mask[tmp_mask == 1] = 1
mask = np.logical_and(result_mask, depth_gt_mask_uint8)
depth[mask==1] = depth_gt[mask==1]
return depth
# def handle_depth(depth, depth_gt_mask):
# depth[depth_gt_mask==1] = 0
# depth_gt_mask_uint8 = np.where(depth < 0.000000001, 255, 0).astype(np.uint8)
# depth_gt_mask_uint8[depth_gt_mask_uint8 != 0] = 1
# depth_uint8 = depth.copy() / depth.max() * 255
# depth_uint8 = np.array(depth_uint8, dtype=np.uint8)
# depth_uint8 = cv2.inpaint(depth_uint8, depth_gt_mask_uint8, 3, cv2.INPAINT_NS)
# depth_uint8 = np.array(depth_uint8, dtype=np.float32) / 255 * depth.max()
# depth[depth_gt_mask==1] = depth_uint8[depth_gt_mask == 1]
# mask_pixel_indices = np.array(np.where(depth_gt_mask == 1)).T # Shape: [#nonzero_pixels x 2]
# dropout_size = int(mask_pixel_indices.shape[0] * 0.005)
# dropout_centers_indices = np.random.choice(mask_pixel_indices.shape[0], size=dropout_size)
# dropout_centers = mask_pixel_indices[dropout_centers_indices, :]
# x_radii = np.random.gamma(3.0, 2.0, size=dropout_size)
# y_radii = np.random.gamma(3.0, 2.0, size=dropout_size)
# angles = np.random.randint(0, 360, size=dropout_size)
# result_mask = np.zeros_like(depth_gt_mask, dtype=np.uint8)
# for i in range(dropout_size):
# center = dropout_centers[i, :]
# x_radius = np.round(x_radii[i]).astype(int)
# y_radius = np.round(y_radii[i]).astype(int)
# angle = angles[i]
# # get ellipse mask
# tmp_mask = np.zeros_like(depth_gt_mask, dtype=np.uint8)
# tmp_mask = cv2.ellipse(tmp_mask, tuple(center[::-1]), (x_radius, y_radius), angle=angle, startAngle=0, endAngle=360, color=1, thickness=-1)
# # update depth and corrupt mask
# result_mask[tmp_mask == 1] = 1
# mask = np.logical_and(np.logical_not(result_mask), depth_gt_mask_uint8)
# depth[mask==1] = 0
return depth
def handle_depth2(depth, depth_gt_mask):
rgb_mask = np.where(depth < 0.000000001, 255, 0).astype(np.uint8)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT,(7,7))
neg_zero_mask_dilated = cv2.dilate(rgb_mask, kernel = kernel)
depth[neg_zero_mask_dilated == 255] = 0.0
contours,hierarch=cv2.findContours(neg_zero_mask_dilated,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_NONE)
for i in range(len(contours)):
area = cv2.contourArea(contours[i]) #计算轮廓所占面积
if area < 320 * 240 // 16: #将area小于阈值区域填充背景色，由于OpenCV读出的是BGR值
cv2.drawContours(depth,[contours[i]],-1, (0), thickness=-1) #原始图片背景BGR值(84,1,68)
depth[neg_zero_mask_dilated == 255] = 0.0
return depth
# def convertPNG(pngfile,outdir):
# # READ THE DEPTH
# im_depth = cv2.imread(pngfile)
# #apply colormap on deoth image(image must be converted to 8-bit per pixel first)
# im_color=cv2.applyColorMap(cv2.convertScaleAbs(im_depth,alpha=15),cv2.COLORMAP_JET)
# #convert to mat png
# im=Image.fromarray(im_color)
# #save image
# im.save(os.path.join(outdir,os.path.basename(pngfile)))
def draw_point_cloud(color, depth, camera_intrinsics, use_mask = False, use_inpainting = True, scale = 1000.0, inpainting_radius = 5, fault_depth_limit = 0.2, epsilon = 0.01):
"""
Given the depth image, return the point cloud in open3d format.
The code is adapted from [graspnet.py] in the [graspnetAPI] repository.
"""
d = depth.copy()
c = color.copy() / 255.0
if use_inpainting:
fault_mask = (d < fault_depth_limit * scale)
d[fault_mask] = 0
inpainting_mask = (np.abs(d) < epsilon * scale).astype(np.uint8)
d = cv2.inpaint(d, inpainting_mask, inpainting_radius, cv2.INPAINT_NS)
fx, fy = camera_intrinsics[0, 0], camera_intrinsics[1, 1]
cx, cy = camera_intrinsics[0, 2], camera_intrinsics[1, 2]
xmap, ymap = np.arange(d.shape[1]), np.arange(d.shape[0])
xmap, ymap = np.meshgrid(xmap, ymap)
points_z = d / scale
points_x = (xmap - cx) / fx * points_z
points_y = (ymap - cy) / fy * points_z
points = np.stack([points_x, points_y, points_z], axis = -1)
if use_mask:
mask = (points_z > 0)
points = points[mask]
c = c[mask]
else:
points = points.reshape((-1, 3))
c = c.reshape((-1, 3))
cloud = o3d.geometry.PointCloud()
cloud.points = o3d.utility.Vector3dVector(points)
cloud.colors = o3d.utility.Vector3dVector(c)
return cloud
inferencer = Inferencer()
# rgb_mask = np.array(Image.open('/home/apollo/TransCG/cleargrasp/cleargrasp-dataset-train/cup-with-waves-train/segmentation-masks/000000000-segmentation-mask.png'), dtype = np.float32)
# rgb = np.array(Image.open('/home/apollo/TransCG/cleargrasp/cleargrasp-dataset-train/cup-with-waves-train/rgb-imgs/000000000-rgb.jpg'), dtype = np.float32)
# depth = exr_loader('/home/apollo/TransCG/cleargrasp/cleargrasp-dataset-train/cup-with-waves-train/depth-imgs-rectified/000000000-depth-rectified.exr', ndim = 1, ndim_representation = ['R'])
# depth_gt = exr_loader('/home/apollo/TransCG/cleargrasp/cleargrasp-dataset-train/cup-with-waves-train/depth-imgs-rectified/000000000-depth-rectified.exr', ndim = 1, ndim_representation = ['R'])
# rgb_mask = np.array(Image.open('/home/apollo/TransCG/cleargrasp/cleargrasp-dataset-test-val/real-test/d415/000000082-mask.png'), dtype = np.float32)
# rgb = np.array(Image.open('/home/apollo/TransCG/cleargrasp/cleargrasp-dataset-test-val/real-test/d415/000000082-transparent-rgb-img.jpg'), dtype = np.float32)
# depth = exr_loader('/home/apollo/TransCG/cleargrasp/cleargrasp-dataset-test-val/real-test/d415/000000082-transparent-depth-img.exr', ndim = 1, ndim_representation = ['R'])
# depth_gt = exr_loader('/home/apollo/TransCG/cleargrasp/cleargrasp-dataset-test-val/real-test/d415/000000082-opaque-depth-img.exr', ndim = 1, ndim_representation = ['R'])
rgb = np.array(Image.open('transcg_data/scene3/1/rgb1.png'), dtype = np.float32)
rgb_mask = np.array(Image.open('transcg_data/scene3/1/depth1-gt-mask.png'), dtype = np.float32)
depth = np.array(Image.open('transcg_data/scene3/1/depth1.png'), dtype = np.float32)
depth_gt = np.array(Image.open('transcg_data/scene3/1/depth1-gt.png'), dtype = np.float32)
depth = depth / 1000
depth_gt = depth_gt / 1000
rgbcopy = rgb.copy()
depth_copy = depth.copy()
rgb_mask_copy = rgb_mask.copy()
res = inferencer.inference(rgb, depth)
cam_intrinsics = np.load('transcg_data/camera_intrinsics/1-camIntrinsics-D435.npy')
depth_gt[np.isnan(depth_gt)] = 0.0
# depth_gt[rgb_mask != 0] = 0
rgb_mask = np.where(depth_gt < 0.000000001, 255, 0).astype(np.uint8)
depth = cv2.resize(depth, (320, 240), interpolation = cv2.INTER_NEAREST)
depth_copy = cv2.resize(depth_copy, (320, 240), interpolation = cv2.INTER_NEAREST)
depth_gt = cv2.resize(depth_gt, (320, 240), interpolation = cv2.INTER_NEAREST)
res = cv2.resize(res, (320, 240), interpolation = cv2.INTER_NEAREST)
rgb_mask = cv2.resize(rgb_mask, (320, 240), interpolation = cv2.INTER_NEAREST).astype(np.uint8)
rgb_mask_copy = cv2.resize(rgb_mask_copy, (320, 240), interpolation = cv2.INTER_NEAREST).astype(np.uint8)
rgb_mask_copy[rgb_mask_copy != 0] = 1
rgb_mask[rgb_mask != 0] = 1
depth_mask = np.where(depth < 0.000000001, 255, 0).astype(np.uint8)
depth_mask[depth_mask != 0] = 1
depth_gt_new = handle_depth(depth_gt.copy(), depth_gt.copy(), rgb_mask_copy)
# depth = handle_depth(depth.copy(), depth_mask)
# depth_gt_mask = dropout_random_ellipses_4corruptmask(rgb_mask_copy, {"ellipse_dropout_mean": 20, "ellipse_gamma_shape": 10.0, "ellipse_gamma_scale": 1.0})
# depth[depth_gt_mask==1] = 0
# depth_gt[rgb_mask==255] = 0
# depth_gt[np.isnan(depth_gt)] = 0.0
# rgb_mask = np.where(depth_gt < 0.000000001, 255, 0).astype(np.uint8)
# rgb_mask[rgb_mask != 0] = 1
# depth_gt = depth_gt / depth.max() * 255
# depth_gt = np.array(depth_gt, dtype=np.uint8)
# depth_gt = cv2.inpaint(depth_gt, rgb_mask, 5, cv2.INPAINT_NS)
# depth_gt = np.array(depth_gt, dtype=np.float32) / 255 * depth.max()
# depth_gt = np.where(depth_gt < 0.3, 0, depth_gt)
# depth_gt = np.where(depth_gt > 1.5, 0, depth_gt)
neg_zero_mask = np.where(depth_gt < 0.0000001)
# zero_mask = np.logical_not(neg_zero_mask)
res[neg_zero_mask] = 0
depth_gt[neg_zero_mask] = 0
depth[neg_zero_mask] = 0
neg_zero_mask = np.where(depth_gt > 5)
res[neg_zero_mask] = 0
depth_gt[neg_zero_mask] = 0
depth[neg_zero_mask] = 0
# depth_gt[neg_zero_mask] = 0
# res = np.clip(res, 0.3, 1.5)
# depth = np.clip(depth, 0.3, 1.5)
# depth_gt = np.clip(depth_gt, 0.3, 1.5)
fig, axs = plt.subplots(2, 2)
tt="hsv"
rgb_1=rgbcopy.astype(np.int8)
axs.flat[0].imshow(rgb_1,cmap=tt)
axs.flat[0].set_title("rgb")
axs.flat[1].imshow(rgb_mask,cmap=tt)
axs.flat[1].set_title("original")
axs.flat[2].imshow(depth,cmap=tt)
axs.flat[2].set_title("model output")
axs.flat[3].imshow(depth_gt,cmap=tt)
axs.flat[3].set_title("groud truth")
plt.show()
# cloud = draw_point_cloud(rgb, res, cam_intrinsics, scale = 1.0)
# cloud_gt = draw_point_cloud(rgb, depth_gt, cam_intrinsics, scale = 1.0)
# frame = o3d.geometry.TriangleMesh.create_coordinate_frame(0.1)
# sphere = o3d.geometry.TriangleMesh.create_sphere(0.002,20).translate([0,0,0.490])
# o3d.visualization.draw_geometries([cloud, cloud_gt, frame, sphere])