[Submitted on 29 Sep 2026 (v1), last revised 23 Oct 2026 (this version, v2)]
Abstract:We benchmarked vision-language models (VLMs) on the decisions annotators take when inspecting electron microscopy images in connectomics: synapse proofreading (presence and polarity) and neuron proofreading (split errors and merge errors). For synapse proofreading, we evaluated 19 open and 2 closed models across various architectures and sizes under zero-shot, four-shot in-context learning and LoRA settings, against specialist models, on datasets constructed by us using public resources. For neuron proofreading, we evaluated 3 open and 2 closed models on the ConnectomeBench2 dataset, with cross-species transfer from fly and mouse to human and zebrafish. Most models were at chance zero-shot; a few examples helped mainly the closed and largest open ones. LoRA on a few thousand labels brought open models level with specialist models. When evaluated on unseen species, the best adapted VLMs outperformed specialist models in identifying merge errors. The project will be publicly available upon acceptance.
Submission history
From: Yicong Li [view email]
[v1] Tue, 29 Sep 2026 01:48:42 UTC (561 KB)
[v2] Fri, 23 Oct 2026 05:15:15 UTC (561 KB)
