ZeroFlow: Scalable Scene Flow via Distillation
Authors: Kyle Vedder, Neehar Peri, Nathaniel Eliot Chodosh, Ishan Khatri, ERIC EATON, Dinesh Jayaraman, Yang Liu, Deva Ramanan, James Hays
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We perform extensive experiments on the Argoverse 2 (Wilson et al., 2021) and Waymo Open (Sun et al., 2020) datasets. We compare to author implementations of NSFP (Li et al., 2021b) and Chodosh et al. (2023), implement Fast Flow3D (Jund et al., 2021) ourselves (no author implementation is available), and use Chodosh et al. (2023)’s implementations for all other baselines. |
| Researcher Affiliation | Collaboration | Kyle Vedder1 Neehar Peri2 Nathaniel Chodosh2 Ishan Khatri3 Eric Eaton1 Dinesh Jayaraman1 Yang Liu4 Deva Ramanan2 James Hays5 1University of Pennsylvania 2Carnegie Mellon University 3Motional 4Lawrence Livermore National Laboratory 5Georgia Tech |
| Pseudocode | Yes | Algorithm 1 Zero Flow |
| Open Source Code | Yes | To facilitate further research, we release our code, trained model weights, and high quality pseudo-labels for the Argoverse 2 and Waymo Open datasets at https://vedder.io/zeroflow. |
| Open Datasets | Yes | We perform extensive experiments on the Argoverse 2 (Wilson et al., 2021) and Waymo Open (Sun et al., 2020) datasets. |
| Dataset Splits | Yes | Table 1: Quantitative results on the Argoverse 2 Sensor validation split using the evaluation protocol from Chodosh et al. (2023). |
| Hardware Specification | Yes | Fast Flow3D (Jund et al., 2021), which uses a Point Pillar-style encoder (Lang et al., 2019), can process 1 million points in under 100 ms on an NVIDIA Tesla P1000 GPU (making it real-time for a 10Hz Li DAR)... Runtimes are collected on an NVIDIA V100 with a batch size of 1 (Peri et al., 2023). |
| Software Dependencies | No | The paper does not specify version numbers for any software dependencies or libraries used in the experiments (e.g., PyTorch, TensorFlow, CUDA). |
| Experiment Setup | No | The paper states: “For details on the exact dataset construction and method hyperparameters, see Supplemental A”. This indicates that detailed experimental setup information, including hyperparameters, is provided in supplementary material and not directly in the main text. |