Thanks for the great work!
The kitti bin only have x,y,z,intensity(not (x, y, z, intensity, channel)), how to augment it with snowfall in your paper?

Hey Martin,

Can you please confirm if we can use "dense_dataset_snow_wet_coupled.yaml" to reproduce result for "Your snow+wet" and "dense_dataset_snow_uniform_gunn_1in10.yaml" to reproduce result for "Your snow"?

Does this mean you only used "lidar_hdl64_strongest" (and not hdl 64 last or vlp32 strongest/last) for training and validating all experiments? Also, the train val splits are given in LiDAR_snow_sim/splits folder but you also have FOV3000last, FOV3000strongest info pkls in your OpenPCDet/data/dense repo. While training for snow/fog sim (both papers), did you ignore frames that have less than 3000 points in camera FOV? If you did, can you please point me where in your code you do that or if 3000 points check is done before or after snow/fog sim?

Your reply will be greatly appreciated. Thanks!

hi Martin,

I have questions regarding Equation 6 in the main paper. I can understand that R^2 comes from the solid angle, transmittance takes into consideration the medium beforehand, I don't understand why it's transmittance squared. Can you elaborate a bit here? Also in the text you mentioned "H C depends on the beam divergence, the overlap of transmitter and receiver described by ξ (R) as well as the transmittance T(R) of the medium through", however, I don't see where does the beam divergence play the role here. I understand beam divergence for two roles: 1. intersect with multiple particles within the cone. 2. adjust the energy that follows Gaussian distribution within the cone. So how does it plays the role in the impulse responses of the optical channel?

Best,
Shengyu

Hello, I have a question of .npy snowflakes file.

I used the snowflake .npy file you provided. I wonder why the number of snowflakes decreases when the value of rainrate is larger.
Shouldn't the number of snowflake particles increase if the rainrate is large?

Thanks.

Hello Martin,
Thanks for your great work. May I know whether there is a tutorial about running the snow simulation on other dataset to create new data?

Hi really appreciate the work that you have shared. I was trying to give it a try but getting an error while cloning the repo

[email protected]: Permission denied (publickey).
fatal: Could not read from remote repository.

even doing the sudo clone i am still getting errors related to access rights

Please make sure you have the correct access rights
and the repository exists.
fatal: clone of '[email protected]:MartinHahner/LISA' into submodule path '/home/ammar/lidar_simulator/LiDAR_snow_sim/lib/LISA' failed
Failed to clone 'lib/LISA'. Retry scheduled
Cloning into '/home/ammar/lidar_simulator/LiDAR_snow_sim/lib/LiDAR_fog_sim'...
[email protected]: Permission denied (publickey).
fatal: Could not read from remote repository.

Please make sure you have the correct access rights
and the repository exists.
fatal: clone of '[email protected]:MartinHahner/LiDAR_fog_sim' into submodule path '/home/ammar/lidar_simulator/LiDAR_snow_sim/lib/LiDAR_fog_sim' failed
Failed to clone 'lib/LiDAR_fog_sim'. Retry scheduled
Cloning into '/home/ammar/lidar_simulator/LiDAR_snow_sim/lib/OpenPCDet'...
[email protected]: Permission denied (publickey).
fatal: Could not read from remote repository.

Please make sure you have the correct access rights
and the repository exists.
fatal: clone of '[email protected]:MartinHahner/OpenPCDet' into submodule path '/home/ammar/lidar_simulator/LiDAR_snow_sim/lib/OpenPCDet' failed

Would appreciate if i can get help

I have a question, in the function “process_single_channel”，it is understandable that pc[j, 4] = 1 or pc[j, 4] = 2 can be defined as a snowflake point, while pc[j, 4] = 0 is a normal point? right?

And then I find in wads datasets and cadcd datasets, The intensity of most snowflake points is 0(zero), So from your simulation, most snowflake point is not 0(zero). Can you explain my question?

I can find nowhere about this equation (Chapter3, eq(2)):

Correction factors are listed for each beam. Did I lose something?

As state in the paper We split the 3916 samples in the snow test set based on the intensity of snowfall into two different subsets, termed light snowfall and heavy snowfall, with 2512 and 1404 samples respectively.

However, I can't find it. Can you opensource its index in the STF?

I noticed that in tools/snowfall/simulation.py there is a beta_0. I suppose it to be reflectivity. Why it's fixed to beta_0 = 1 * 10 ** -6 / PI

Hello，thank you for your excellent work, it is exactly what I need.
I have seen the https://github.com/MartinHahner/LiDAR_fog_sim and https://github.com/SysCV/LiDAR_snow_sim. In the fog simulation model, there is a fog_simulation.py to input clear pointcloud files and output fog pointcloud files. While, is there any way to achieve similar functionality in the snow+wet simulation model, or could you please tell me how to achieve this? Hoping for your reply.

Thanks for your great work! I wanna know what is the purpose of this file called precompute.py, because I do not see any corresponding note.

Excuse me，I'm trying to save the point cloud data after adding rain and snow noise by running the code.
But in the actual code, I can only click on the button to show the point cloud for the next frame one by one.
Could you tell me how to save data in bulk without having to click on it one at a time?
Thank you very much!

Hey Martin,

I am training pointrcnn on dense dataset clear weather split. However it breaks here.

The problem is that the number of field of view points becomes less than half of 16384 (due to mask_points_and_boxes_outside_range function).
Note: 16384 is the number of points sampled for pointrcnn by "sample_points".

For example:
If after masking points outside range, the len(points) is 5000, then
choice = np.arange(5000)
np.random.choice(, 16384-5000, replace=False) breaks because np.choice cannot sample number of points (i.e. 11384) larger than the number of points in choice (i.e. 5000) without replacing.

How did you manage to overcome this issue during training?

Thank you for the help!

Hello, I used the code provided, but there exists gap between the results I reproduced and the results the paper provided.

This is the result I reproduced in PV-RCNN model.

This is the result the paper reported in PV-RCNN model.

And I used the model you have provided experiments/snow+wet/pc_rcnn.pth, and the result is:

Looking forward to your reply!

Only a add our fog, no add our snow+wet

Hi,
in file: tools/snowfall/simulation.py, line123 - 126:

        if channel in [53, 55, 56, 58]:
            max_intensity = 230
        else:
            max_intensity = 255

Why did not max_intensity come from calibration file?

Thank you!

Hello,

Are snow and wet augmentations only available for the DENSE dataset?

Hi,

Great work!

I'm wondering what machine (gpu) you use and training time?

Excuse me, where is the path to save the point cloud file after adding snowflake simulation noise? Thank you！

Hi @MartinHahner

The paper seems to only consider the scatter points (raindrops) generated between the object point and the origin(similar to LISA), but raindrops(scatter points) generated in open spaces are not considered.
So, I would like to know whether raindrop points can be still generated if they are in a very open area?

Hey Martin,

Which classes from Seeing through fog dataset did you group into the kitti "Car", "Pedestrian", "Cyclist" classes? Your OpenPCDet dense_dataset.py assumes that the groundtruth labels for dense dataset are "Car", "Pedestrian", "Cyclist".

Thanks

Dear Martin, I met a bug when I was running precompute.py. The reason for the problem is that the variable called ground_distances is empty and the error happens on this file called 2018-02-09_19-10-30_00000.bin. I do not know how to solve this problem because I haven't changed your other code except the corresponding path. Please give me some advice.

gunn: 63%|██████▎ | 2194/3469 [00:47<00:27, 46.29it/s] Traceback (most recent call last): File "C:/Users/Roland/Desktop/LiDAR_snow_sim-main/tools/snowfall/precompute.py", line 107, in <module> beam_divergence=float(np.degrees(3e-3))) File "C:\Users\Roland\Desktop\LiDAR_snow_sim-main\tools\snowfall\simulation.py", line 466, in augment p = np.polyfit(ground_distances, adaptive_noise_threshold, 2) File "<__array_function__ internals>", line 6, in polyfit File "C:\Users\Roland\miniconda3\envs\point_rcnn\lib\site-packages\numpy\lib\polynomial.py", line 630, in polyfit raise TypeError("expected non-empty vector for x") TypeError: expected non-empty vector for x

hi, nice repo!

i wonder know when will the code be public?

thanks!

Hi, Martin!
Thank you so much for your work. I'm confused about the training process. I am debugging at the dense_data.py using the openpcdet in the project. And there seems something wrong with the function 'def get_infos', so I am wondering how to arrrange the STF dataset under' openpcdet. data. dense' just using hdl64&strongest.

Thank you for your reply!

Thank you for your excellent paper in this field. Will the training code in the paper be made public? If it does, it will help us greatly in our scientific research.