Depth-discriminative Metric Learning for Monocular 3D Object Detection
Authors: Wonhyeok Choi, Mingyu Shin, Sunghoon Im
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The broad applicability of our method is demonstrated through experiments that show improvements in overall performance when integrated into various baselines. The results show that our method consistently improves the performance of various baselines by 25.27% and 4.54% on average across KITTI and Waymo, respectively. |
| Researcher Affiliation | Academia | Wonhyeok Choi Mingyu Shin Sunghoon Im DGIST, Daegu, Korea {smu06117, alsrb4446, sunghoonim}@dgist.ac.kr |
| Pseudocode | Yes | Algorithm 1 (K, B, ϵ)-Quasi-isometric loss Input: h, (ui, vi), zi, where i {1, 2, ..., n}, n is the number of objects in a batch. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing source code for the described methodology or a link to a code repository. |
| Open Datasets | Yes | The KITTI dataset [12] consists of 7,481 training images and 7,518 test images for official KITTI 3D object detection evaluation and contains three categories: Car, Pedestrian, and Cyclist. The Waymo dataset [32] is a recently released dataset comprising 798 training sequences and 202 validation sequences, with four categories: Vehicles, Pedestrians, Cyclists, and Signs. |
| Dataset Splits | Yes | For additional experiments, we follow [23], which splits the training images into 3,712 and 3,769 images for training and validation sets, respectively. We use the split reported in [28], including 52,386 training and 39,848 validation images, to evaluate performance on the Waymo dataset. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments, such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | The paper mentions using Center Net-based frameworks and refers to various papers for experimental setups but does not list specific software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions). |
| Experiment Setup | Yes | As an exception, we set all batch sizes to 16 for the KITTI and Waymo benchmarks. For the (K, B, ϵ)-quasi-isometric loss, we use K = 1.5, B = 0.5, ϵ = 10.0, d1( , ) = | , |, d2( , ) = , 2 for all experiments. |