Research map / 3D vision / 3D generation
Panoptic Compositional Feature Field for Editable Scene Rendering With Network-Inferred Labels via Metric Learning
Xinhua Cheng, Yanmin Wu, Mengxi Jia, Qian Wang, Jian Zhang
This paper tried a common change: Larger batch size
Train PCFF with a larger batch size · Replica · PSNR: 32.28 -> 32.77 Improved
What 785 accepted papers report for this change
- 469 improved
- 107 no clear effect
- 48 mixed
- 161 hurt
- Predicting the Emergence of Induction Heads in Language Model Pretraining ICML 2026 · increase batch size (fixed context 1024) · Natural data (CC100) · induction head emergence point (updates): rho=-1.0, p<0.001Improved
- When RL Meets Adaptive Speculative Training: A Unified Training-Serving System ICML 2026 · Increase batch size from 1 to 32 · MiniMax M2.1 FP8 · speedup: 1.57x -> 1.25xHurt
- Untied Ulysses: Memory-Efficient Context Parallelism via Headwise Chunking ICML 2026 · increase per-device batch size to 10 using freed memory · Llama 3-8B, 128K · throughput (TPS): 2464.10 (Ulysses) -> 2526.77 (UPipe)Improved
See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.
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 |
|---|---|---|---|
| Train PCFF with network-inferred panoptic labels using instance quadruplet metric learning | PCFF · ScanNet target-object extraction | IoU: 0.486 -> 0.891 | Improved |
| Use semantic-appearance hierarchical learning (SAHL) (ablation) | PCFF · ScanNet 0192 | PSNR: 26.84 -> 26.86 | Improved |
| Use semantic-appearance hierarchical learning (SAHL) (ablation) | PCFF · Replica office4 | PSNR: 31.52 -> 31.68 | Improved |
| Add semantic-guided regional refinement (SGRR) to SAHL (ablation) | PCFF · ScanNet 0192 | PSNR: 26.86 -> 27.02 | Improved |
| Add semantic-guided regional refinement (SGRR) to SAHL (ablation) | PCFF · Replica office4 | PSNR: 31.68 -> 31.86 | Improved |
| Add instance quadruplet metric learning (IQML) to SAHL and SGRR (ablation) | PCFF · ScanNet 0192 | PSNR: 27.02 -> 26.96 | Hurt |
| Add instance quadruplet metric learning (IQML) to SAHL and SGRR (ablation) | PCFF · Replica office4 | PSNR: 31.86 -> 31.76 | Hurt |
| Replace the proposed quadruplet scheme with triplet metric loss (ablation) | PCFF · ScanNet 0192 | PSNR: 26.89 vs. 26.96 | Hurt |
| Replace the proposed quadruplet scheme with triplet metric loss (ablation) | PCFF · Replica office4 | PSNR: 31.65 vs. 31.76 | Hurt |
| Use the inverse quadruplet margin scheme instead of smaller margins for hard negatives (ablation) | PCFF · ScanNet 0192 | PSNR: 26.88 vs. 26.96 | Hurt |
| Use the inverse quadruplet margin scheme instead of smaller margins for hard negatives (ablation) | PCFF · Replica office4 | PSNR: 31.67 vs. 31.76 | Hurt |
| Train PCFF with a larger batch size | PCFF · Replica | PSNR: 32.28 -> 32.77 | Improved |
Closest papers by meaning
- Learning Object-Compositional Neural Radiance Field for Editable Scene Rendering (ICCV 2021 · similarity 0.89)
- Learning Unified Decompositional and Compositional NeRF for Editable Novel View Synthesis (ICCV 2023 · similarity 0.88)
- Panoptic Neural Fields: A Semantic Object-Aware Neural Scene Representation (CVPR 2022 · similarity 0.88)
- Compositional Scene Representation Learning via Reconstruction: A Survey (TPAMI 2023 · similarity 0.87)
- SCION: Scene Composition via Instanced Neural Primitives (NeurIPS 2026 · similarity 0.85)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.85)
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.
