fix: use object args for vision processor

This commit is contained in:
SysVis AI 2025-12-28 20:30:57 +07:00
parent 023cc3264c
commit cc74967a54

View file

@ -91,8 +91,10 @@ export class VisionService {
// Task: Detailed Captioning is best for understanding diagrams // Task: Detailed Captioning is best for understanding diagrams
const text = '<MORE_DETAILED_CAPTION>'; const text = '<MORE_DETAILED_CAPTION>';
// Pass image as an array to ensure it's iterable for transformers.js preprocessing // Pass arguments as object to avoid positional ambiguity
const inputs = await this.processor([image], text); // Florence-2 processor typically expects 'images' and 'text'
if (!this.processor) throw new Error('Processor is undefined');
const inputs = await this.processor({ text, images: [image] });
const generatedIds = await this.model.generate({ const generatedIds = await this.model.generate({
...inputs, ...inputs,