Research map / Detection and segmentation / Open-set and few-shot detection
Neural-Sim: Learning to Generate Training Data with NeRF
Yunhao Ge, Harkirat Behl, Jiashu Xu, Suriya Gunasekar, Neel Joshi, Yale Song, Xin Wang, Laurent Itti, Vibhav Vineet
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add bi-level optimization of rendering parameters (NSO vs NS) | Neural-Sim · YCB-synthetic (21 objects) | mAP: 68.4 -> 82.1 | Improved |
| use NSO instead of Auto-Sim | Auto-Sim · YCB-synthetic (21 objects) | mAP: 69.3 -> 82.1 | Improved |
| add NSO to optimize pose distribution when train/test pose gap exists | Neural-Sim · YCB-synthetic pitcher and driller | average precision: almost 30% and 10% | Improved |
| use differentiable NSO instead of REINFORCE-based LTS/Auto-Sim for pose | LTS and Auto-Sim · YCB-synthetic pitcher | average precision: almost 34% and 11% | Improved |
| add NSO illumination optimization with NeRF-w | Neural-Sim · YCB-synthetic illumination | average precision: 16% and 15% | Improved |
| add NSO on real YCB-in-the-wild images | Neural-Sim · YCB-in-the-wild pitcher | accuracy: almost 60% | Improved |
| add optimization (NSO) over NS in few-shot YCB-Video | train(pre)+ours · YCB-Video 10-shot and 25-shot | mAP: 3.9 -> 4.9; 4.6 -> 4.9 | Improved |
| add NS synthetic data to few-shot real data | Only YCBV-train · YCB-Video few-shot | mAP: 0.45 -> 3.9; 0.49 -> 4.6 | Improved |
| add NSO optimization to NS in zero-shot | train(pre)+ours · YCB-Video 0-shot | mAP: 2.3 -> 4.5 | Improved |
| add optimization to combined real+synthetic training | train(com)+ours · YCB-Video limited data | mAP: 36.7 -> 37.4; 43.6 -> 44.9; 47.9 -> 48.9 | Improved |
| use NSO instead of Auto-Sim on combined limited data | Auto-Sim (com) · YCB-Video limited data (10%) | mAP: 48.3 -> 48.9 | Improved |
| gap between train and test pose or zoom distributions | object detector · YCB-synthetic pitcher and driller | average precision: almost 30% and 10% drop | Hurt |
Closest papers by meaning
- D-NeRF: Neural Radiance Fields for Dynamic Scenes (CVPR 2021 · similarity 0.89)
- Neural Radiance Fields From Sparse RGB-D Images for High-Quality View Synthesis (TPAMI 2022 · similarity 0.89)
- AligNeRF: High-Fidelity Neural Radiance Fields via Alignment-Aware Training (CVPR 2023 · similarity 0.89)
- Mask-Based Modeling for Neural Radiance Fields (ICLR 2024 · similarity 0.88)
- Language-driven Object Fusion into Neural Radiance Fields with Pose-Conditioned Dataset Updates (CVPR 2024 · similarity 0.88)
- CF-NeRF: Camera Parameter Free Neural Radiance Fields with Incremental Learning (AAAI 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
