Research map / Architectures and efficiency / RNNs and normalization
Pip-Stereo: Progressive Iterations Pruner for Iterative Optimization based Stereo Matching
Jintu Zheng, Qizhe Liu, Huangxin Xu, Zhuojie Chen
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 monocular depth prior transfer (MPT) (ablation) | Selective-IGEV · SceneFlow test | EPE: 0.44 -> 0.38 (32 iters) | Improved |
| add supernet search for student block allocation (ablation) | PipStereo MPT · SceneFlow test | EPE: 0.40 -> 0.38 | Improved |
| add MPT when using a single inference iteration (ablation) | Selective-IGEV · SceneFlow test | EPE: 0.64 -> 0.56 | Improved |
| add PIP finetuning when pruning to 1 iteration (ablation) | PipStereo-i32 · SceneFlow test | EPE: 0.56 -> 0.45 | Improved |
| prune iterations from 32 to 16 with PIP (ablation) | PipStereo · SceneFlow test | EPE: 0.38 -> 0.38 | No clear effect |
| prune iterations further (16 to 1) with PIP (ablation) | PipStereo · SceneFlow test | EPE: 0.38 -> 0.45 | Hurt |
| add PIP to single-iteration FoundationStereo (ablation) | FoundationStereo · ETH3D | Bad-1 (Noc): 0.94 -> 0.70 | Improved |
| add PIP to single-iteration RAFT-Stereo (ablation) | RAFT-Stereo · SceneFlow test | EPE: 3.65 -> 2.16 | Mixed |
| add FlashGRU to PIP-pruned model (8 iters) (ablation) | PipStereo-p2 · SceneFlow latency | latency (s): 1.02 -> 0.76 | Improved |
| add FlashGRU to PIP-pruned RAFT-Stereo (4 iters) (ablation) | Raft-p3 · SceneFlow latency | latency (s): 1.07 -> 0.69 | Improved |
| add FlashGRU to ConvGRU at 2K resolution | Selective-IGEV update block · RTX 4090 1280x2944 | RNN latency: 122.4ms -> 16.8ms (x7.28) | Improved |
| add FlashGRU at low resolution | Selective-IGEV update block · RTX 4090 320x736 | RNN latency: 28.8ms -> 14.8ms (x1.94) | Improved |
Closest papers by meaning
- DeepPruner: Learning Efficient Stereo Matching via Differentiable PatchMatch (ICCV 2019 · similarity 0.89)
- StereoNet: Guided Hierarchical Refinement for Real-Time Edge-Aware Depth Prediction (ECCV 2018 · similarity 0.88)
- Open-World Stereo Video Matching with Deep RNN (ECCV 2018 · similarity 0.88)
- AdaStereo: An Efficient Domain-Adaptive Stereo Matching Approach (IJCV 2022 · similarity 0.88)
- Any-Stereo: Arbitrary Scale Disparity Estimation for Iterative Stereo Matching (AAAI 2024 · similarity 0.87)
- IINet: Implicit Intra-inter Information Fusion for Real-Time Stereo Matching (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.
