Infoblock alpha
med/imaging IB-arxiv-2607.28148-f1-976-samples measured

Under 5-fold cross-validation on 976 expert-annotated TCM tongue images (TongueDx2), the best model (ConvNeXt-Tiny with restrained augmentation and weak-group ensemble) reached weighted-F1 of 0.6625.

Active Evidence state is recorded explicitly and may be revised by a later event.

Observed quantities

weighted_f1 0.6625 · n_samples 976 · cv_folds 5

01

Primary source

  1. arxiv · 2607.28148 · v1 · abstract
Title
What Makes Deep Learning Work for Traditional Chinese Medicine Tongue Diagnosis? A Comprehensive Ablation Study
Authors
Longxia Gao, Linan Wang, Yuhe Han, Junze Geng, Meng Zhang, Hanqing Zhao
Version
v1
Check
Support quote found word for word in the arXiv abstract v1, 2026-10-08
Supporting quote from the source
The best 976-sample model achieved weighted-F1 of 0.6625 using ConvNeXt-Tiny with restrained augmentation and weak-group ensemble

02

Boundary of this block

No additional public limitation was recorded.

03

Connections

Outgoing

No outgoing public edges.

Incoming

No incoming public edges.