{"id":17594923,"url":"https://github.com/andreaconti/lidar-confidence","last_synced_at":"2025-10-15T09:55:33.683Z","repository":{"id":73012627,"uuid":"543065573","full_name":"andreaconti/lidar-confidence","owner":"andreaconti","description":"Effective unsupervised framework aimed at explicitly addressing confidence estimation of LiDAR sparse depth maps","archived":false,"fork":false,"pushed_at":"2023-04-23T11:54:51.000Z","size":2077,"stargazers_count":21,"open_issues_count":1,"forks_count":2,"subscribers_count":1,"default_branch":"master","last_synced_at":"2025-05-12T14:25:49.259Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Jupyter Notebook","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/andreaconti.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2022-09-29T10:43:25.000Z","updated_at":"2023-09-26T03:07:22.000Z","dependencies_parsed_at":"2024-10-23T02:17:46.762Z","dependency_job_id":null,"html_url":"https://github.com/andreaconti/lidar-confidence","commit_stats":null,"previous_names":[],"tags_count":1,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/andreaconti%2Flidar-confidence","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/andreaconti%2Flidar-confidence/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/andreaconti%2Flidar-confidence/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/andreaconti%2Flidar-confidence/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/andreaconti","download_url":"https://codeload.github.com/andreaconti/lidar-confidence/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":253754428,"owners_count":21958861,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2024-10-22T07:24:32.099Z","updated_at":"2025-10-15T09:55:28.635Z","avatar_url":"https://github.com/andreaconti.png","language":"Jupyter Notebook","funding_links":[],"categories":[],"sub_categories":[],"readme":"# [Unsupervised confidence for LiDAR depth maps and applications](https://arxiv.org/pdf/2210.03118.pdf)\n\n\u003cp\u003e\n\u003cdiv align=\"center\"\u003e\n    \u003ca href=\"https://andreaconti.github.io\"\u003eAndrea Conti\u003c/a\u003e\n    \u0026middot;\n    \u003ca href=\"https://mattpoggi.github.io\"\u003eMatteo Poggi\u003c/a\u003e\n    \u0026middot;\n    \u003ca href=\"https://filippoaleotti.github.io/website\"\u003eFilippo Aleotti\u003c/a\u003e\n    \u0026middot;\n    \u003ca href=\"http://vision.deis.unibo.it/~smatt/Site/Home.html\"\u003eStefano Mattoccia\u003c/a\u003e\n\u003c/div\u003e\n\u003cdiv align=\"center\"\u003e\n    \u003ca href=\"https://arxiv.org/pdf/2210.03118.pdf\"\u003e[Arxiv]\u003c/a\u003e\n    \u003ca href=\"https://github.com/andreaconti/lidar-confidence/blob/master/torchhub-example.ipynb\"\u003e[Demo]\u003c/a\u003e\n\u003c/div\u003e\n\u003c/p\u003e\n\nLiDAR (Light Detection And Ranging) sensors are rapidly spreading in autonomous driving applications thanks to their long-range effectiveness and their dropping costs, leading  to  an  increasing  number  of  applications  exploiting  sparse  depth  data  comingfrom LiDAR devices.  However, when coupled with RGB cameras and projected over images, LiDAR depth maps expose several outliers due to noise and, more frequently, dis-occlusions between the sensor itself and the RGB cameras, resulting in large errors on the inputs to all the applications that process such depth maps.\n\nThis repository collects the code and the final models used to study in our [paper](https://arxiv.org/pdf/2210.03118.pdf) the issue of estimating the confidence of the LiDAR depthmaps by leveraging a deep learning framework. The dataset used is a subset of the KITTI dataset.\n\n|                                              Image                                               |                                           Lidar Depth                                            |                                                 Lidar Depth Filtered                                                 |\n| :----------------------------------------------------------------------------------------------: | :----------------------------------------------------------------------------------------------: | :------------------------------------------------------------------------------------------------------------------: |\n| ![image](https://github.com/andreaconti/lidar-confidence/blob/master/resources/teaser/image.png) | ![lidar](https://github.com/andreaconti/lidar-confidence/blob/master/resources/teaser/lidar.png) | ![lidar confidence](https://github.com/andreaconti/lidar-confidence/blob/master/resources/teaser/lidar_filtered.png) |\n\n## Citation\n\n```\n@inproceedings{aconti2022lidarconf,\n  title={Unsupervised confidence for LiDAR depth maps and applications},\n  author={Conti, Andrea and Poggi, Matteo and Aleotti, Filippo and Mattoccia, Stefano},\n  booktitle={IEEE/RSJ International Conference on Intelligent Robots and Systems},\n  note={IROS},\n  year={2022}\n}\n```\n\n## Reproduce the experiments\n\nTo build a working environment use [conda](https://docs.conda.io/en/latest/) and\n[dvc](https://dvc.org), they will take care of downloading the dependencies and\ndata. If conda takes too much time too build the environment please try to use\n[mamba](https://mamba.readthedocs.io/en/latest/user_guide/mamba.html).\n\n```bash\n$ # download the project\n$ git clone https://github.com/andreaconti/lidar-confidence\n$ cd lidar-confidence\n\n$ # setup of the virtualenv, note that conda installs dvc\n$ conda env create -f environment.yml\n$ conda activate lidar-confidence\n$ pip install -e .\n\n$ # download the data\n$ # NOTE: the first time you call dvc pull it will ask you to login on\n$ #       google drive, asking to open your browser, it will require to\n$ #       redirect to localhost:8080 to complete the procedure. So if\n$ #       you are working on a remote machine please apply ssh port\n$ #       forwarding like ssh -L 8080:localhost:8080 \u003cusername\u003e@\u003cmachine\u003e\n$ dvc pull  # or dvc fetch \u0026 dvc checkout\n\n$ # now you have both data and code and you can run the experiments\n$ dvc repro -R experiments\n```\n\nWhen you reproduce the training procedure the experiments will try to connect to\n[wandb](https://wandb.ai) to log results, you can control and even disable wandb\nentirely using environment variables listed\n[here](https://docs.wandb.ai/library/environment-variables).\n\nAt the end of the experiments all the metrics will be logged in .json format in\nthe folder [results](https://github.com/andreaconti/lidar-confidence/tree/master/results)\nanyway.\n\n## Download a pretrained model\n\nYou can use [dvc](https://dvc.org) itself to download the pretrained models\ndirectly from this project, see [here](https://dvc.org/doc/start/data-and-model-access).\nNote that all the pre-trained models can be found\nin [results](https://github.com/andreaconti/lidar-confidence/tree/master/results), for\ninstance you can download the pre-trained unsupervised model using:\n\n```bash\n$ dvc get https://github.com/andreaconti/lidar-confidence results/lce/model.pth\n```\n\nAll models are in [TorchScript](https://pytorch.org/docs/stable/jit.html) format\nthus you can simply download the file and load it as below, without the need of\nthe model source code:\n\n```python\nIn [1]: !dvc get https://github.com/andreaconti/lidar-confidence results/lce/model.pth\nIn [2]: import torch\nIn [3]: model = torch.jit.load(\"model.pth\")\n```\n\n## TorchHub Interface\n\nFinally, we provide a simple TorchHub interface to load the pretrained model and the 142 split dataset. A minimal example to reproduce results can be found [here](https://github.com/andreaconti/lidar-confidence/blob/master/torchhub-example.ipynb)\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fandreaconti%2Flidar-confidence","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fandreaconti%2Flidar-confidence","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fandreaconti%2Flidar-confidence/lists"}