clean code
This commit is contained in:
Kai-46
parent
a51403c562
commit
a581e48614
3 changed files with 1 additions and 220 deletions
131
data_verifier.py
131
data_verifier.py
|
|
@ -1,131 +0,0 @@
|
||||||
import cv2
|
|
||||||
import numpy as np
|
|
||||||
|
|
||||||
|
|
||||||
## pip install opencv-python=3.4.2.17 opencv-contrib-python==3.4.2.17
|
|
||||||
|
|
||||||
|
|
||||||
def skew(x):
|
|
||||||
return np.array([[0, -x[2], x[1]],
|
|
||||||
[x[2], 0, -x[0]],
|
|
||||||
[-x[1], x[0], 0]])
|
|
||||||
|
|
||||||
|
|
||||||
def two_view_geometry(intrinsics1, extrinsics1, intrinsics2, extrinsics2):
|
|
||||||
'''
|
|
||||||
:param intrinsics1: 4 by 4 matrix
|
|
||||||
:param extrinsics1: 4 by 4 W2C matrix
|
|
||||||
:param intrinsics2: 4 by 4 matrix
|
|
||||||
:param extrinsics2: 4 by 4 W2C matrix
|
|
||||||
:return:
|
|
||||||
'''
|
|
||||||
relative_pose = extrinsics2.dot(np.linalg.inv(extrinsics1))
|
|
||||||
R = relative_pose[:3, :3]
|
|
||||||
T = relative_pose[:3, 3]
|
|
||||||
tx = skew(T)
|
|
||||||
E = np.dot(tx, R)
|
|
||||||
F = np.linalg.inv(intrinsics2[:3, :3]).T.dot(E).dot(np.linalg.inv(intrinsics1[:3, :3]))
|
|
||||||
|
|
||||||
return E, F, relative_pose
|
|
||||||
|
|
||||||
|
|
||||||
def drawpointslines(img1, img2, lines1, pts2, color):
|
|
||||||
'''
|
|
||||||
draw corresponding epilines on img1 for the points in img2
|
|
||||||
'''
|
|
||||||
|
|
||||||
r, c = img1.shape
|
|
||||||
img1 = cv2.cvtColor(img1, cv2.COLOR_GRAY2BGR)
|
|
||||||
img2 = cv2.cvtColor(img2, cv2.COLOR_GRAY2BGR)
|
|
||||||
for r, pt2, cl in zip(lines1, pts2, color):
|
|
||||||
x0, y0 = map(int, [0, -r[2]/r[1]])
|
|
||||||
x1, y1 = map(int, [c, -(r[2]+r[0]*c)/r[1]])
|
|
||||||
cl = tuple(cl.tolist())
|
|
||||||
img1 = cv2.line(img1, (x0,y0), (x1,y1), cl, 1)
|
|
||||||
img2 = cv2.circle(img2, tuple(pt2), 5, cl, -1)
|
|
||||||
return img1, img2
|
|
||||||
|
|
||||||
|
|
||||||
def epipolar(coord1, F, img1, img2):
|
|
||||||
# compute epipole
|
|
||||||
pts1 = coord1.astype(int).T
|
|
||||||
color = np.random.randint(0, high=255, size=(len(pts1), 3))
|
|
||||||
# Find epilines corresponding to points in left image (first image) and
|
|
||||||
# drawing its lines on right image
|
|
||||||
lines2 = cv2.computeCorrespondEpilines(pts1.reshape(-1,1,2), 1,F)
|
|
||||||
lines2 = lines2.reshape(-1,3)
|
|
||||||
img3, img4 = drawpointslines(img2,img1,lines2,pts1,color)
|
|
||||||
## print(img3.shape)
|
|
||||||
## print(np.concatenate((img4, img3)).shape)
|
|
||||||
## cv2.imwrite('vis.png', np.concatenate((img4, img3), axis=1))
|
|
||||||
|
|
||||||
return np.concatenate((img4, img3), axis=1)
|
|
||||||
|
|
||||||
|
|
||||||
def verify_data(img1, img2, intrinsics1, extrinsics1, intrinsics2, extrinsics2):
|
|
||||||
img1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
|
|
||||||
img2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)
|
|
||||||
|
|
||||||
E, F, relative_pose = two_view_geometry(intrinsics1, extrinsics1,
|
|
||||||
intrinsics2, extrinsics2)
|
|
||||||
|
|
||||||
# sift = cv2.xfeatures2d.SIFT_create(nfeatures=20)
|
|
||||||
# kp1 = sift.detect(img1, mask=None)
|
|
||||||
# coord1 = np.array([[kp.pt[0], kp.pt[1]] for kp in kp1]).T
|
|
||||||
|
|
||||||
# Initiate ORB detector
|
|
||||||
orb = cv2.ORB_create()
|
|
||||||
# find the keypoints with ORB
|
|
||||||
kp1 = orb.detect(img1, None)
|
|
||||||
coord1 = np.array([[kp.pt[0], kp.pt[1]] for kp in kp1[:20]]).T
|
|
||||||
return epipolar(coord1, F, img1, img2)
|
|
||||||
|
|
||||||
|
|
||||||
if __name__ == '__main__':
|
|
||||||
from data_loader import load_data
|
|
||||||
from run_nerf import config_parser
|
|
||||||
from nerf_sample_ray import parse_camera
|
|
||||||
import os
|
|
||||||
|
|
||||||
parser = config_parser()
|
|
||||||
args = parser.parse_args()
|
|
||||||
print(args)
|
|
||||||
|
|
||||||
data = load_data(args.datadir, args.scene, testskip=1)
|
|
||||||
|
|
||||||
all_imgs = data['images']
|
|
||||||
all_cameras = data['cameras']
|
|
||||||
all_intrinsics = []
|
|
||||||
all_extrinsics = [] # W2C
|
|
||||||
for i in range(all_cameras.shape[0]):
|
|
||||||
W, H, intrinsics, extrinsics = parse_camera(all_cameras[i])
|
|
||||||
all_intrinsics.append(intrinsics)
|
|
||||||
all_extrinsics.append(np.linalg.inv(extrinsics))
|
|
||||||
|
|
||||||
#### arbitrarily select 10 pairs of images to verify pose
|
|
||||||
out_dir = os.path.join(args.basedir, args.expname, 'data_verify')
|
|
||||||
print(out_dir)
|
|
||||||
os.makedirs(out_dir, exist_ok=True)
|
|
||||||
|
|
||||||
def calc_angles(c2w_1, c2w_2):
|
|
||||||
c1 = c2w_1[:3, 3:4]
|
|
||||||
c2 = c2w_2[:3, 3:4]
|
|
||||||
|
|
||||||
c1 = c1 / np.linalg.norm(c1)
|
|
||||||
c2 = c2 / np.linalg.norm(c2)
|
|
||||||
return np.rad2deg(np.arccos(np.dot(c1.T, c2)))
|
|
||||||
|
|
||||||
images_verify = []
|
|
||||||
for i in range(10):
|
|
||||||
while True:
|
|
||||||
idx1, idx2 = np.random.choice(len(all_imgs), (2,), replace=False)
|
|
||||||
|
|
||||||
angle = calc_angles(np.linalg.inv(all_extrinsics[idx1]),
|
|
||||||
np.linalg.inv(all_extrinsics[idx2]))
|
|
||||||
if angle > 5. and angle < 10.:
|
|
||||||
break
|
|
||||||
|
|
||||||
im = verify_data(np.uint8(all_imgs[idx1]*255.), np.uint8(all_imgs[idx2]*255.),
|
|
||||||
all_intrinsics[idx1], all_extrinsics[idx1],
|
|
||||||
all_intrinsics[idx2], all_extrinsics[idx2])
|
|
||||||
cv2.imwrite(os.path.join(out_dir, '{:03d}.png'.format(i)), im)
|
|
||||||
|
|
@ -7,15 +7,6 @@ import imageio
|
||||||
########################################################################################################################
|
########################################################################################################################
|
||||||
# ray batch sampling
|
# ray batch sampling
|
||||||
########################################################################################################################
|
########################################################################################################################
|
||||||
|
|
||||||
def parse_camera(params):
|
|
||||||
H, W = params[:2]
|
|
||||||
intrinsics = params[2:18].reshape((4, 4))
|
|
||||||
c2w = params[18:34].reshape((4, 4))
|
|
||||||
|
|
||||||
return int(W), int(H), intrinsics.astype(np.float32), c2w.astype(np.float32)
|
|
||||||
|
|
||||||
|
|
||||||
def get_rays_single_image(H, W, intrinsics, c2w):
|
def get_rays_single_image(H, W, intrinsics, c2w):
|
||||||
'''
|
'''
|
||||||
:param H: image height
|
:param H: image height
|
||||||
|
|
|
||||||
81
utils.py
81
utils.py
|
|
@ -8,9 +8,7 @@ HUGE_NUMBER = 1e10
|
||||||
TINY_NUMBER = 1e-6 # float32 only has 7 decimal digits precision
|
TINY_NUMBER = 1e-6 # float32 only has 7 decimal digits precision
|
||||||
|
|
||||||
|
|
||||||
# Misc utils
|
# misc utils
|
||||||
# work on tensors
|
|
||||||
# img2mse = lambda x, y: torch.mean((x - y) * (x - y))
|
|
||||||
def img2mse(x, y, mask=None):
|
def img2mse(x, y, mask=None):
|
||||||
if mask is None:
|
if mask is None:
|
||||||
return torch.mean((x - y) * (x - y))
|
return torch.mean((x - y) * (x - y))
|
||||||
|
|
@ -32,13 +30,6 @@ to8b = lambda x: (255 * np.clip(x, 0, 1)).astype(np.uint8)
|
||||||
# gray2rgb = lambda x: np.tile(x[:,:,np.newaxis], (1, 1, 3))
|
# gray2rgb = lambda x: np.tile(x[:,:,np.newaxis], (1, 1, 3))
|
||||||
mse2psnr = lambda x: -10. * np.log(x+TINY_NUMBER) / np.log(10.)
|
mse2psnr = lambda x: -10. * np.log(x+TINY_NUMBER) / np.log(10.)
|
||||||
|
|
||||||
#
|
|
||||||
# def normalize(x):
|
|
||||||
# x_min, x_max = np.percentile(x, (0.5, 99.5))
|
|
||||||
# x = np.clip(x, x_min, x_max)
|
|
||||||
# x = (x - x_min) / (x_max - x_min)
|
|
||||||
# return x
|
|
||||||
|
|
||||||
|
|
||||||
########################################################################################################################
|
########################################################################################################################
|
||||||
#
|
#
|
||||||
|
|
@ -51,15 +42,6 @@ import cv2
|
||||||
|
|
||||||
|
|
||||||
def get_vertical_colorbar(h, vmin, vmax, cmap_name='jet', label=None):
|
def get_vertical_colorbar(h, vmin, vmax, cmap_name='jet', label=None):
|
||||||
'''
|
|
||||||
:param w: pixels
|
|
||||||
:param h: pixels
|
|
||||||
:param vmin: min value
|
|
||||||
:param vmax: max value
|
|
||||||
:param cmap_name:
|
|
||||||
:param label
|
|
||||||
:return:
|
|
||||||
'''
|
|
||||||
fig = Figure(figsize=(1.2, 8), dpi=100)
|
fig = Figure(figsize=(1.2, 8), dpi=100)
|
||||||
fig.subplots_adjust(right=1.5)
|
fig.subplots_adjust(right=1.5)
|
||||||
canvas = FigureCanvasAgg(fig)
|
canvas = FigureCanvasAgg(fig)
|
||||||
|
|
@ -86,9 +68,6 @@ def get_vertical_colorbar(h, vmin, vmax, cmap_name='jet', label=None):
|
||||||
|
|
||||||
fig.tight_layout()
|
fig.tight_layout()
|
||||||
|
|
||||||
# # debug
|
|
||||||
# fig.savefig("debug3.png")
|
|
||||||
|
|
||||||
canvas.draw()
|
canvas.draw()
|
||||||
s, (width, height) = canvas.print_to_buffer()
|
s, (width, height) = canvas.print_to_buffer()
|
||||||
|
|
||||||
|
|
@ -102,35 +81,6 @@ def get_vertical_colorbar(h, vmin, vmax, cmap_name='jet', label=None):
|
||||||
return im
|
return im
|
||||||
|
|
||||||
|
|
||||||
# def colorize_np(x, cmap_name='jet', append_cbar=False):
|
|
||||||
# vmin = x.min()
|
|
||||||
# vmax = x.max() + TINY_NUMBER
|
|
||||||
# x = (x - vmin) / (vmax - vmin)
|
|
||||||
# # x = np.clip(x, 0., 1.)
|
|
||||||
|
|
||||||
# cmap = cm.get_cmap(cmap_name)
|
|
||||||
# x_new = cmap(x)[:, :, :3]
|
|
||||||
|
|
||||||
# cbar = get_vertical_colorbar(h=x.shape[0], vmin=vmin, vmax=vmax, cmap_name=cmap_name)
|
|
||||||
|
|
||||||
# if append_cbar:
|
|
||||||
# x_new = np.concatenate((x_new, np.zeros_like(x_new[:, :5, :]), cbar), axis=1)
|
|
||||||
# return x_new
|
|
||||||
# else:
|
|
||||||
# return x_new, cbar
|
|
||||||
|
|
||||||
|
|
||||||
# # tensor
|
|
||||||
# def colorize(x, cmap_name='jet', append_cbar=False):
|
|
||||||
# x = x.numpy()
|
|
||||||
# x, cbar = colorize_np(x, cmap_name)
|
|
||||||
|
|
||||||
# if append_cbar:
|
|
||||||
# x = np.concatenate((x, np.zeros_like(x[:, :5, :]), cbar), axis=1)
|
|
||||||
|
|
||||||
# x = torch.from_numpy(x)
|
|
||||||
# return x
|
|
||||||
|
|
||||||
def colorize_np(x, cmap_name='jet', mask=None, append_cbar=False):
|
def colorize_np(x, cmap_name='jet', mask=None, append_cbar=False):
|
||||||
if mask is not None:
|
if mask is not None:
|
||||||
# vmin, vmax = np.percentile(x[mask], (1, 99))
|
# vmin, vmax = np.percentile(x[mask], (1, 99))
|
||||||
|
|
@ -175,32 +125,3 @@ def colorize(x, cmap_name='jet', append_cbar=False, mask=None):
|
||||||
|
|
||||||
x = torch.from_numpy(x)
|
x = torch.from_numpy(x)
|
||||||
return x
|
return x
|
||||||
|
|
||||||
|
|
||||||
if __name__ == '__main__':
|
|
||||||
# # cbar = get_vertical_colorbar(h=512, vmin=0.1, vmax=5, cmap_name='jet')
|
|
||||||
# # cbar = cbar[:, :, :3]
|
|
||||||
# import imageio
|
|
||||||
#
|
|
||||||
# # imageio.imwrite('./debug.png', cbar)
|
|
||||||
#
|
|
||||||
# x = torch.rand(512, 512)
|
|
||||||
# x = colorize(x, append_cbar=True)
|
|
||||||
#
|
|
||||||
# x = np.uint8(x.numpy() * 255.)
|
|
||||||
#
|
|
||||||
# import imageio
|
|
||||||
# imageio.imwrite('./debug.png', x)
|
|
||||||
|
|
||||||
import os
|
|
||||||
import imageio
|
|
||||||
|
|
||||||
img_dir = '/phoenix/S7/kz298/latest_work/nerf/logs/dtu_scan9_3_nearfar/renderonly_train_200001'
|
|
||||||
|
|
||||||
all_imgs = []
|
|
||||||
for item in sorted(os.listdir(img_dir)):
|
|
||||||
if item[-4:] == '.png':
|
|
||||||
fpath = os.path.join(img_dir, item)
|
|
||||||
all_imgs.append(imageio.imread(fpath))
|
|
||||||
|
|
||||||
imageio.mimwrite(os.path.join(img_dir, 'video.mp4'), all_imgs, fps=3, quality=8)
|
|
||||||
|
|
|
||||||
Loading…
Reference in a new issue