Hi all, I am wondering if there is any friend has download the dataset recently? I followed the tutorial to register in DENSE dataset webpage and get the download link via the email. However, seems like the file in the link has been removed. Please help me to find the correct method for getting the dataset:) Thanks.

First of all, thank you for sharing a great dataset.

I have a question. The size of x,y,z of the sensor is 32x400, and can we think of 32 as the number of ring channels of LiDAR? Then, I don't know how the 400 came out, can you tell me?

In addition, why is the intensity value of labeled points zero?

I'll wait for your answer. Thank you.

I noticed that some of the files in the dataset have the same name, except one has a suffix _2, as an example: LidarImage_000000001.hdf5 and LidarImage_000000001_2.hdf5. What does that mean? Are those two files describing the same point in time but one contains the strongest and the other the last returns of the lidar? If yes, which is which?

Also, why do these pairs exist only for a part of the dataset and not all measurements?

Thanks in advance!

Hi

I'm wondering if you've released the denoised output for the chamber data used in the paper Pixel-Accurate Depth Evaluation in Realistic Driving Scenarios (https://ieeexplore.ieee.org/abstract/document/8885465/). The images in this paper look similar to the other paper. Please let me know if there is correspondence and which split I should refer to.

Thanks!

Hi, Thanks for your amazing work! The Adverse Weather has been a troubling question in the 3D field. Experiments from your paper give reasonable results for these scenes, which made me eager to try this task. Would you please release the dataset, or give us a schedule of planned releases? Any help would be greatly appreciated!

Hi Thanks for your great work !

I recently downloaded your dataset but I came across some detailed problem when trying to re-implement your work

How to project your point cloud data to the image plane so there is no calibration file ?

Thanks Best Regards

Hi,

I'm trying to replicate experiment 2 from the paper where about 31k road samples were used. First of all, the uploaded dataset contains only 27462 road samples, so I assume the remaining ~4k are taken from the LiLaNet dataset.

Since that one isn't public, is it possible to publish the chosen selection so that I can take that from the full dataset if I get access to it or to request the subset from the LiLaNet authors?

Furthermore, the uploaded dataset contains only road samples in training directories. Which samples were used for evaluation?

Lastly, where do the ~103k samples in experiment 3 come from? Is it based on the ~34k clear frames where each was augmented with both fog an rain? If yes, where do the additional ~3k frames compared to experiment 2 come from?

Thanks in advance!

In the paper it mentions that 4 realistic static setups are represented in the data from the climate chamber. I presume that the names of the folder that contain Static1, Static2, Static3, Static4 refer to this, however I noted that the train_01 folder only contains Static1 data, train_02 Static4, the test_01 folder only contains Static2 and the val_01 Static3. Is there any reasons for this? Shouldn't the training data contain data from all 4 static setups?

What are the differences between train_01 and train_02? The train_road data is also splited into 2, while being not labeled.

Could you explain why there is this separation and what is the correspondence for the _01 and _02 split?

In DENSE datasets page it mentions the following:

"For the challenging task of lidar point cloud de-noising, we rely on the Pixel Accurate Depth Benchmark and the Seeing Through Fog(STF) dataset recorded under adverse weather conditions like heavy rain or dense fog. In particular, we use the point clouds from a Velodyne VLP32c lidar sensor."

Could you further explain please?

In particular, my questions are the following:

1. Is the data exclusively from those 2 sources (mentioned in the Dense webpage), except the train_road_01 and train_road_02 folders which contain the road data you collected and is not labeled, or does it add some data? This is unclear, especially since the dataset from this repository contain 6.6k clear weather frames (train, test, val folders) but the STF(Seeing Through Fog) dataset mentions using only 5.5k clear weather frames, for the real world recording, which is not used. The climate chamber data it uses contains only 364 clear weather frames according to this paper (TABLE I). So, does this mean that the other 6.3k clear weather frames data comes from the Pixel-Accurate Depth Evaluation in Realistic Driving Scenarios dataset?
2. What part of the data is being used from those papers. From #13 you suggest to use the STF road data data, but in the DENSE datasets page it mentions that the Seeing Through Fog dataset is being used, so I presume that only the climate chamber data of the STF dataset is included.
3. Is the data from the climate chamber in STF and ** "Pixel-Accurate Depth Evaluation in Realistic Driving Scenarios" ***different from each other, the exact same data, or does it contain some overlap?
4. Of the data used from both papers/datasets, is there any change to the data from both papers, or does it contain exactly the same data, in terms of quantity, diversity, distribution?

Essentially, I am confused with the different datasets available in DENSE webpage, and to avoid repetition I am trying to understand overlaps in the data being used. The Dense webpage contains the following datasets to download:

• cnn_denoising (the one refering to this github repo)
• FogChamberDataset
• SeingThroughFog
• pixel_accurate_depth_benchmark

Since all these datasets are referred or mention in the paper, Dense webpage, github issues could you explain me, or point to where should I look, to solve this confusion?

Sorry for the long question xD

First of all, thank you for this excellent job. I have some questions about the final output of the network, is the final output a 32*400 matrix, the same as the labels read from the dataset, and the elements are also 0 and 100,101.102? Or is it a label manipulation or is it a matrix with the number of channels representing categories like Lilanet? Thank you。

Hi @rheinzler Thanks for your great work ! In the paper "CNN-based Lidar Point Cloud De-Noising in Adverse Weather", all of the data is static, but there are two dynamic scene inputs at the beginning of README.md.

my questions are the following: 1.Where can I download the previous input data or similar dynamic continuous point cloud data with rain or fog? could you point to where should I look and download the data? 2.is that the input data of two dynamic scenes also in the image plane with a dimension of 32x400?

Hello @rheinzler, I am trying to reproduce your results I have some doubts about your training and testing data.

In the training set we have the weather chamber data with labels [100: clear, 101: rain, 102: fog] and the unlabeled data from the road scenes [0: unlabeled]. During training, how do you treat the unlabeled data? Do your remap them to the clear class [100] or you ignore them during the loss calculation?

Similarly, in the evaluation section you mention that:

"Note, all evaluations are based on the test data set from experiment 2, which contains autolabeled annotations and road data without fog, rain or augmentation."

However the classes taken into consideration during the evaluation are only [100: clear, 101: rain, 102: fog]. Did you only evaluate on the weather chamber data or did you remap the unlabeled to the clear class?

Hello, First of all thank you for publishing the data, which is extremely helpful for researchers.

In #8 it was noted that the clear weather is the only one that contains both the strongest and latest return of the LiDAR, however it was not specified in the clear weather which label was be used for the strongest and latest return. Since other weather conditions use the strongest return only, I assume that in the clear weather data, the data with _2 represents the latest return and the normal files represent the strongest. Is that correct?

Hi, everyone! My issue is similar to #4 I followed the tutorial to register in DENSE dataset webpage and got the download link via the email. I followed the link and was able to navigate over directories. But when I try to download any file I face the problem that file has 0 b size.

Seems, like all files were removed from the cloud. Please, help me to acquire access to the dataset. Thanks!