martinhahner / lidar_fog_sim Goto Github PK
View Code? Open in Web Editor NEWLiDAR fog simulation
Home Page: https://www.trace.ethz.ch/lidar_fog_sim
License: Other
LiDAR fog simulation
Home Page: https://www.trace.ethz.ch/lidar_fog_sim
License: Other
Thank you for your excellent work.
In fact, it is similar to issue "the wrong result of the fog\u simulation.py using the Kitti data. \7". I directly run the "fog_simulation.py" file on Kitti and waymo datasets. The results are similar to issue 7. It seems that almost all points are regarded as fog noise.
I want to know whether it needs to be modified due to the intensity of the data or whether it is impossible to directly run fog_ simulation. Instead, I need to run pointcloud_viewer.py (doesn't sound reasonable)
Looking forward to your reply
Hi Martin,
Thank you for providing this elegant and powerful repo and I really appreciate your brilliant work!
I'm trying to reproduce the training pipeline on Seeing Through Fog Dataset. In the experimental part of the paper, there are clear weather baseline and "strongest ∩ last filter" baseline. I want to know whether clear weather baseline is evaluated on strongest return or last return of dense fog test split. I don't seem to find a description in the paper or in the supplementary materials, so sorry if I missed it.
Thank you in advance!
Thank you for your excellent contribution of this work!There is a problem when I run the fog_simulation.py to generate the fog by using the kitti dataset.
The 1st picture is the original kitti data.
The 2nd picture is generated by the fog_simulation.py.
I wonder how can i solve the problem,thank you.
Hello
First of all, thank you for these codes, they've helped me a lot.
I am a newbie to 3D object detection, I saw the demo.py code that you wrote in OpenPCDet, I have used it and got as an output the data_dict_idx.pkl and pred_dicts_idx.pkl .
Now I am trying to visualize these outputs using your pointcloud_viewer but when i choose the custom directory where i put the demo.py output, nothing shows up, I was wondering if you could help me
Thanks for your excellent work. I am hoping to use the simulation to generate some synthetic foggy data, e.g. on KITTI. Hence, I am wondering how much does the simulated data match with real-world foggy data? But I found only some minor qualitive discussion is represented in the paper, no quantitive discussion.
Can I only use the dataset in the introduction of the code, what if I want to add this fog simulation to my dataset?
Hello, I have read your paper of "Fog Simulation on Real LiDAR Point Clouds for 3D Object Detection in Adverse Weather," and I have some questions about it. Specifically, I used the STF dataset and divided the clear data into training, testing, and validation sets in a 7:2:1 ratio. I trained the PointPillar and PV-RCNN++ models on the clear training dataset and tested them on the clear validation set based on openpcdet. However, the results are not satisfactory, and I am unable to achieve the impressive results mentioned in your article. This issue has been bothering me for a long time. I apologize for the inconvenience, and I seek your assistance. Thank you.
some results as fellows:
Pedestrian [email protected], 0.50, 0.50:
bbox AP:15.5895, 15.2841, 15.5391
bev AP:11.0431, 11.1163, 11.3700
3d AP:10.4871, 10.5488, 10.2467
aos AP:11.37, 11.10, 11.26
Pedestrian [email protected], 0.50, 0.50:
bbox AP:9.3038, 9.4138, 9.3635
bev AP:3.7221, 3.5240, 3.7916
3d AP:2.8358, 2.6212, 2.5333
aos AP:5.32, 5.35, 5.32
Pedestrian [email protected], 0.25, 0.25:
bbox AP:15.5895, 15.2841, 15.5391
bev AP:15.1044, 15.0117, 15.5118
3d AP:14.8881, 14.7912, 15.3198
aos AP:11.37, 11.10, 11.26
Pedestrian [email protected], 0.25, 0.25:
bbox AP:9.3038, 9.4138, 9.3635
bev AP:8.7162, 8.9601, 9.1860
3d AP:8.4221, 8.4356, 8.8460
aos AP:5.32, 5.35, 5.32
PassengerCar [email protected], 0.70, 0.70:
bbox AP:0.0000, 0.0000, 0.0000
bev AP:0.0000, 0.0000, 0.0000
3d AP:0.0000, 0.0000, 0.0000
aos AP:0.00, 0.00, 0.00
PassengerCar [email protected], 0.70, 0.70:
bbox AP:0.0000, 0.0000, 0.0000
bev AP:0.0000, 0.0000, 0.0000
3d AP:0.0000, 0.0000, 0.0000
aos AP:0.00, 0.00, 0.00
PassengerCar [email protected], 0.50, 0.50:
bbox AP:0.0000, 0.0000, 0.0000
bev AP:0.0000, 0.0000, 0.0000
3d AP:0.0000, 0.0000, 0.0000
aos AP:0.00, 0.00, 0.00
PassengerCar [email protected], 0.50, 0.50:
bbox AP:0.0000, 0.0000, 0.0000
bev AP:0.0000, 0.0000, 0.0000
3d AP:0.0000, 0.0000, 0.0000
aos AP:0.00, 0.00, 0.00
RidableVehicle [email protected], 0.50, 0.50:
bbox AP:0.0000, 0.0000, 0.0000
bev AP:0.0000, 0.0000, 0.0000
3d AP:0.0000, 0.0000, 0.0000
aos AP:0.00, 0.00, 0.00
RidableVehicle [email protected], 0.50, 0.50:
bbox AP:0.0000, 0.0000, 0.0000
bev AP:0.0000, 0.0000, 0.0000
3d AP:0.0000, 0.0000, 0.0000
aos AP:0.00, 0.00, 0.00
RidableVehicle [email protected], 0.25, 0.25:
bbox AP:0.0000, 0.0000, 0.0000
bev AP:0.0000, 0.0000, 0.0000
3d AP:0.0000, 0.0000, 0.0000
aos AP:0.00, 0.00, 0.00
RidableVehicle [email protected], 0.25, 0.25:
bbox AP:0.0000, 0.0000, 0.0000
bev AP:0.0000, 0.0000, 0.0000
3d AP:0.0000, 0.0000, 0.0000
aos AP:0.00, 0.00, 0.00
Hi! I came from OpenPCDet issues channel and found your amazing work!
But I have a problem about loading the result.pkl with kitti velodyne data. I can't visualize the predicted bbox in the point cloud.
Is there anything wrong with my file path with the pickle?
Thanks a lot in advance! The interface is really great.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.