AI Paper Atlas

Research map / Video / Video-language understanding

BANZ-FS: BANZSL Fingerspelling Dataset

Xin Shen, Yan Ke, Xinyu Wang, Xin Yu

ICLR 2026 · Video › Video-language understanding · approach: Video LLMs with fine-grained understanding

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 changedWhereMeasuredEffect
use character-level tokenization (ByT5) instead of subword tokenization (T5)Transformer translation model · BANZ-FS FSR-ContextLetter Accuracy (%): 10.2% -> 25.8%Improved
train on the combined Full set instead of News onlyHandReader · BANZ-FS IFSR (test Full)Letter Accuracy (%): 60.8 -> 75.4Improved
train in-domain on Web instead of Lab data, tested on WebHandReader · BANZ-FS IFSR (test Web)Letter Accuracy (%): 32.6 -> 55.0Improved
train on Full instead of Lab only, tested on LabHandReader · BANZ-FS IFSR (test Lab)Letter Accuracy (%): 93.1 -> 86.7Hurt
train on Full instead of Web only, tested on WebFS-PoseNet · BANZ-FS IFSR (test Web)Letter Accuracy (%): 51.4 -> 70.1Improved
use multimodal RGB + 3D pose input (HandReader) instead of pose-only FS-PoseNetIFSR model · BANZ-FS IFSR (train Full, test Full)Letter Accuracy (%): 74.7 -> 75.4Improved
use frame-level BIO tagging with pose cues (SL-Seg) instead of proposal-based Modified R-C3Dfingerspelling detector · BANZ-FS FSD (train Full, test Full)AP@IoU0.5: 48.8 -> 66.9Improved
add recognition requirement (FSD-R) on top of detectionSL-Seg · BANZ-FS FSD to FSD-RAP: 66.9 -> 53.5Hurt
train detector on Lab only instead of Web only, tested on WebSL-Seg · BANZ-FS FSD (test Web)AP@IoU0.5: 21.5 -> 40.4Hurt
use pose input instead of RGB inputSL-Luong · Auslan-Daily News V2 (single-person)ROUGE: 15.47 -> 21.71Improved
crop the signer (single-person) instead of full frame (multi-person)GFSLT-VLP · Auslan-Daily News V2ROUGE: 21.87 -> 26.24Improved
crop the signer (single-person) instead of full frame (multi-person)GASLT · Auslan-Daily News V2ROUGE: 20.78 -> 23.94Improved

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