Can not run Euroc datasets

[lihanlun]

I have used VIODE datasets to test the dynaVINS, and got results as better as the paper write. However, when I use the Euroc datasets to test the dynaVINS, I find that, the dynaVINS cannot get a correct results, the results is not convergence. Have you tried other datasets except other open datasets except the paper used.

[ForLoveandPeace]

Hi, I have tried it and it worked well. Maybe you need to check the config of euroc.

[lihanlun]

I have checked it, and can not find the problem. My euroc config file as flows, could you help me check it?

Extrinsic parameter between IMU and Camera.

estimate_extrinsic: 0 # 0 Have an accurate extrinsic parameters. We will trust the following imu^R_cam, imu^T_cam, don't change it.
# 1 Have an initial guess about extrinsic parameters. We will optimize around your initial guess.

body_T_cam0: !!opencv-matrix
rows: 4
cols: 4
dt: d
data: [0.0148655429818, -0.999880929698, 0.00414029679422, -0.0216401454975,
0.999557249008, 0.0149672133247, 0.025715529948, -0.064676986768,
-0.0257744366974, 0.00375618835797, 0.999660727178, 0.00981073058949,
0.0, 0.0, 0.0, 1.0]

body_T_cam1: !!opencv-matrix
rows: 4
cols: 4
dt: d
data: [0.0125552670891, -0.999755099723, 0.0182237714554, -0.0198435579556,
0.999598781151, 0.0130119051815, 0.0251588363115, 0.0453689425024,
-0.0253898008918, 0.0179005838253, 0.999517347078, 0.00786212447038,
0.0, 0.0, 0.0, 1.0]

#Multiple thread support
multiple_thread: 1

#feature traker paprameters
max_cnt: 150 # max feature number in feature tracking
min_dist: 15 # min distance between two features
freq: 10 # frequence (Hz) of publish tracking result. At least 10Hz for good estimation. If set 0, the frequence will be same as raw image
F_threshold: 2.0 # ransac threshold (pixel)
max_depth: 10.0 # max estimated depth (m)
show_track: 1
show_image_feat_weight: 1

publish tracking image as topic

flow_back: 1 # perform forward and backward optical flow to improve feature tracking accuracy

#dynaVINS parameters
dyna_on: true # do not change it to false
regularization_lambda: 2.0
momentum_on: true
momentum_lambda: 0.2
alternating_converge: 0.9
margin_feature_thresh: 0.1

#optimization parameters
max_solver_time: 0.3 # max solver itration time (ms), to guarantee real time
max_num_iterations: 10 # max solver itrations, to guarantee real time
keyframe_parallax: 10.0 # keyframe selection threshold (pixel)

#imu parameters The more accurate parameters you provide, the better performance
acc_n: 0.2 # accelerometer measurement noise standard deviation.
gyr_n: 0.05 # gyroscope measurement noise standard deviation.
acc_w: 0.02 # accelerometer bias random work noise standard deviation.
gyr_w: 4.0e-5 # gyroscope bias random work noise standard deviation.
g_norm: 9.81007 # gravity magnitude

#unsynchronization parameters
estimate_td: 0 # online estimate time offset between camera and imu
td: 0.0 # initial value of time offset. unit: s. readed image clock + td = real image clock (IMU clock)

[lihanlun]

cam0 config file as fellows:
model_type: PINHOLE
camera_name: camera
image_width: 752
image_height: 480
distortion_parameters:
k1: -2.9545645106987750e-01
k2: 8.6623215640186171e-02
p1: 2.0132892276082517e-06
p2: 1.3924531371276508e-05
projection_parameters:
fx: 4.6115862106007575e+02
fy: 4.5975286598073296e+02
cx: 3.6265929181685937e+02
cy: 2.4852105668448124e+02

[lihanlun]

cam1 config file as fellows:
model_type: PINHOLE
camera_name: camera
image_width: 752
image_height: 480
distortion_parameters:
k1: -2.9294124381930947e-01
k2: 8.4798002331543665e-02
p1: -2.9984646536002372e-04
p2: 3.0028216325237329e-04
projection_parameters:
fx: 4.6009781682258682e+02
fy: 4.5890983492218902e+02
cx: 3.7314916359808268e+02
cy: 2.5440734973672119e+02

[lihanlun]

Or, Could you share your euroc config files?

[ForLoveandPeace]

I have used evo to evaluate dynaVINS on Machine Hall dataset. As result, VINS-Fusion's performance is better than dynaVINS's.
Actually, your config file is almost same with mine. Could you show more information about the error?

[lihanlun]

Thank you for your share information. When I use euroc MH01, at the beginning it is OK, then show the wrong traj ( traj as straight line). Now I find that, I should change some code:
while(costDiff<DYN_ALT_CONV && iteration<5)
{
optimizeVIO(true,costDiff);
optimizeVIO(false,costDiff);
std::cout<<"COSTDIFF:"<<costDiff<<std::endl;
iteration++;
}
To

while(costDiff<DYN_ALT_CONV && iteration<5)
{
optimizeVIO(false,costDiff);
optimizeVIO(true,costDiff);
std::cout<<"COSTDIFF:"<<costDiff<<std::endl;
iteration++;
}

It can get a correct traj, but worse than vins-fusion.

[lihanlun]

I think the code should fix feature_weight first then fix others and optimize feature_weight. A very big problem is that the results get worse.

[zhipengak]

I have the same problem, DynaVins can't get the reasonable result with Euroc dataset.
Also I try it with my dataset I acquired from my chassis, bad trajectory I got.
But both Euroc dataset and my dataset I can run well in VINS-fusions!

[ForLoveandPeace]

I have the same problem, DynaVins can't get the reasonable result with Euroc dataset.
Also I try it with my dataset I acquired from my chassis, bad trajectory I got.
But both Euroc dataset and my dataset I can run well in VINS-fusions!

Hi, could you show more details? Thanks!

[zhipengak]

Problem solved!
I replace the yaml "cam0_pinhole.yaml" and "cam1_pinhole.yaml" with "cam0_mei.yaml" and "cam0_mei.yaml", it work and output the right trajectory! I dont know why "cam0_pinhole.yaml" and "cam1_pinhole.yaml" are not work, but glad this code can work with "cam0_mei.yaml" and "cam0_mei.yaml"!
Thank you for share this code very much!

[lihanlun]

Problem solved! I replace the yaml "cam0_pinhole.yaml" and "cam1_pinhole.yaml" with "cam0_mei.yaml" and "cam0_mei.yaml", it work and output the right trajectory! I dont know why "cam0_pinhole.yaml" and "cam1_pinhole.yaml" are not work, but glad this code can work with "cam0_mei.yaml" and "cam0_mei.yaml"! Thank you for share this code very much!

I have tried this method，but the result is still wrose than vins-fusion. Have you evaluated the results on Euroc datasets.