ViT shrinks 54.5x to 6MB for on-device crop disease detection
jm_alexia · x · 2026-09-07
An arXiv paper proposes a unified ViT compression framework combining Hessian-balanced adaptive block pruning (H-BAC), quantization, and attention-based knowledge distillation for on-device deployment in resource-constrained agriculture.
- Model shrinks from 327.42MB to 6.01MB (54.5x) while retaining 95.13% of FP32 accuracy on a cross-village chilli disease test
- A same-size INT8 student trained directly reaches 94.87%
- Commenter notes: deployment win is clear, but the distillation gain remains unshown
More from Research
- Spotting reward hacking in benchmarks is far harder than you think, says AI2 researcher — soldni · 2026-10-26
- RCT: ChatGPT temporarily boosts consultants' data science skills, gains vanish without it — EricTopol · 2026-10-26
- FocusVLA routes robot camera views by predicted interaction intent across task stages — iscas-cloud · 2026-10-26
- Transformers' biggest audio overhaul: unified AudioProcessor architecture — LysandreJik · 2026-10-26
- Free 195-page primer: The Little Book of Generative AI Foundations — caglar_ee · 2026-10-26
- SPADER from Zhejiang and Peking University Tops All Four Multi-Answer QA Benchmarks, Accepted at EMNLP — jiqizhixin · 2026-10-26