cenekalbl / mvus Goto Github PK

Hello,

I ran the code on your dataset 4. I observed that the mean error of each camera became nan after the 2nd BA with 3 cameras.

Here are the inputs I used :

1. Config file: I used the example config file in the README.md.
2. Dataset: I used dataset 4 in https://github.com/CenekAlbl/drone-tracking-datasets
3. Code: I changed line 1190 in reconstruction/common.py to usecols=(0,1,2) so that it fits the data format in the dataset.

Here is the environment:
Python 3.6 on Ubuntu 16.04 with Numpy 1.19.2, Scipy 1.5.2 and Opencv 4.4

Here are the outputs:
Input data are loaded successfully, a scene is created.

The given corresponding frames are directly exploited as temporal synchronization

----------------- Bundle Adjustment with 2 cameras -----------------

Mean error of each camera before BA: [125.5008 130.0244]
Number of BA parameters is 5094
Doing BA with 2 cameras...

Mean error of each camera after first BA: [203.2418 310.8224]
Number of BA parameters is 5094
Doing BA with 2 cameras...

Mean error of each camera after second BA: [ 7.5155 349.8521]

Total time: 0:03:27.236546

----------------- Bundle Adjustment with 3 cameras -----------------

Mean error of each camera before BA: [ 5.0291 349.8246 1900.3164]
Number of BA parameters is 5184
Doing BA with 3 cameras...

Mean error of each camera after first BA: [ 333.0888 349.8245 1874.2762]
Number of BA parameters is 5184
Doing BA with 3 cameras...

/playpen/haojiang/code/mvus/mvus_env/lib/python3.6/site-packages/numpy/core/fromnumeric.py:3373: RuntimeWarning: Mean of empty slice.
out=out, **kwargs)
/playpen/haojiang/code/mvus/mvus_env/lib/python3.6/site-packages/numpy/core/_methods.py:170: RuntimeWarning: invalid value encountered in double_scalars
ret = ret.dtype.type(ret / rcount)

Mean error of each camera after second BA: [ nan nan 4.9613]

Total time: 0:05:35.977872

----------------- Bundle Adjustment with 4 cameras -----------------

Mean error of each camera before BA: [ nan nan 4.9632 523.0124]
Number of BA parameters is 5151
Doing BA with 4 cameras...

Mean error of each camera after first BA: [ nan nan 4.9632 523.0124]
Number of BA parameters is 5151
Doing BA with 4 cameras...

Mean error of each camera after second BA: [ nan nan 4.9632 nan]

Total time: 0:06:06.315165

----------------- Bundle Adjustment with 5 cameras -----------------

Mean error of each camera before BA: [ nan nan 4.9632e+00 nan 6.1845e+03]
Number of BA parameters is 5025
Doing BA with 5 cameras...

Mean error of each camera after first BA: [ nan nan 6.4515 nan 1714.2829]
Number of BA parameters is 5025
Doing BA with 5 cameras...

Mean error of each camera after second BA: [ nan nan 3.9593 nan nan]

Total time: 0:07:22.998962

----------------- Bundle Adjustment with 6 cameras -----------------

Mean error of each camera before BA: [ nan nan 3.9588 nan nan 3008.8645]
Number of BA parameters is 5013
Doing BA with 6 cameras...

Mean error of each camera after first BA: [ nan nan 3.9489 nan nan 2336.0663]
Traceback (most recent call last):
File "/playpen/haojiang/code/mvus/multiviewunsynch/main.py", line 56, in
flight.remove_outliers(flight.sequence[:cam_temp],thres=flight.settings['thres_outlier'])
File "/playpen/haojiang/code/mvus/multiviewunsynch/reconstruction/common.py", line 714, in remove_outliers
self.detection_to_global(i)
File "/playpen/haojiang/code/mvus/multiviewunsynch/reconstruction/common.py", line 126, in detection_to_global
detect = self.cameras[i].undist_point(self.detections[i][1:]) if self.settings['undist_points'] else self.detections[i][1:]
File "/playpen/haojiang/code/mvus/multiviewunsynch/reconstruction/common.py", line 1154, in undist_point
dst = cv2.undistortPoints(src, self.K, self.d)
cv2.error: OpenCV(4.4.0) /tmp/pip-req-build-6amqbhlx/opencv/modules/calib3d/src/undistort.dispatch.cpp:276: error: (-215:Assertion failed) CV_IS_MAT(_src) && CV_IS_MAT(_dst) && (_src->rows == 1 || _src->cols == 1) && (_dst->rows == 1 || _dst->cols == 1) && _src->cols + _src->rows - 1 == _dst->rows + _dst->cols - 1 && (CV_MAT_TYPE(_src->type) == CV_32FC2 || CV_MAT_TYPE(_src->type) == CV_64FC2) && (CV_MAT_TYPE(_dst->type) == CV_32FC2 || CV_MAT_TYPE(_dst->type) == CV_64FC2) in function 'cvUndistortPointsInternal'

Thanks in advance,

Hao

Hi,

Thanks for the great work!

One sample config file is provided for dataset 4, I was wondering if you could provide example configs for other datasets? Or more specifically, the corresponding_frames for each camera in other datasets since it is an essential set of numbers and seem to be missing..

Traceback (most recent call last):
File "W:\mvus\multiviewunsynch\main.py", line 33, in
flight.init_traj(error=flight.settings['thres_Fmatix'])
File "W:\mvus\multiviewunsynch\reconstruction\common.py", line 197, in init_traj
F,inlier = ep.computeFundamentalMat(d1[1:],d2[1:],error=error)
File "W:\mvus\multiviewunsynch\reconstruction\epipolar.py", line 116, in computeFundamentalMat
return F, mask.reshape(-1,)
AttributeError: 'NoneType' object has no attribute 'reshape'

error happens when the value of "thres_Fmatix" set to be too small

Hi I just read your paper and are looking for a bit more insight in what "of the shelf tracking" that were used. Could you please elaborate on how you performed that step?

Hello Cenek,
Thank you for the amazing work you shared here. I'm very interesting in your research and I have a lot of questions.
But, my English is very poor,so I think it should be better to discuss my questions with Jintong Li, the first author of the paper, you know, in Chinese.
I had sent an e-mail to [email protected], but it seems that a wrong address.
So, could you tell me how can I contact with him?
Yours,
Jianzheng Liu