AI Paper Atlas

Research map / Point clouds / Segmentation and detection

Fast Point Transformer

Chunghyun Park, Yoonwoo Jeong, Minsu Cho, Jaesik Park

CVPR 2022 · Point clouds › Segmentation and detection · approach: Scene completion and instance segmentation · 217 citations (OpenAlex)

Check 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 changedWhereMeasuredEffect
use Fast Point Transformer instead of Point Transformer3D semantic segmentation · S3DIS Area 5 testLatency (seconds): 18.07 -> 0.14Improved
use Fast Point Transformer instead of MinkowskiNet42†3D semantic segmentation · S3DIS Area 5 testmIoU (%): 67.1 -> 68.5Improved
use Fast Point Transformer instead of MinkowskiNet423D semantic segmentation · ScanNet validation, voxel size 2cmmIoU (%): 72.2 -> 72.1No clear effect
replace MinkowskiNet† backbone with Fast Point Transformer backboneVoteNet · ScanNet validationmAP@0.25: 55.3 -> 59.1Improved
add centroid-aware voxelization encoding (ablation)Fast Point Transformer · ScanNet validation, voxel size 10cmmIoU (%): 59.1 -> 61.3Improved
add centroid-aware devoxelization encoding (ablation)Fast Point Transformer · ScanNet validation, voxel size 10cmmIoU (%): 61.3 -> 62.7Improved
add continuous positional encoding δ_abs (ablation)Fast Point Transformer · ScanNet validation, voxel size 10cmmIoU (%): 62.7 -> 65.3Improved
use cosine attention instead of softmax attention (ablation)Fast Point Transformer · ScanNet validation, voxel size 10cmmIoU (%): 61.0 -> 65.3Improved
add query-key similarity to cosine attention (ablation)Fast Point Transformer · ScanNet validation, voxel size 10cmmIoU (%): 65.3 -> 62.1Hurt
increase local window size from 3 to 5 (ablation)Fast Point Transformer · ScanNet validation, voxel size 10cmmIoU (%): 65.3 -> 62.4Hurt
increase local window size from 3 to 7 (ablation)Fast Point Transformer · ScanNet validation, voxel size 10cmmIoU (%): 65.3 -> 61.9Hurt
decompose continuous positional encodings (ablation)Fast Point Transformer · ScanNet validation, voxel size 2cmPeak memory usage (GB): 9.52 -> 3.61Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy