BBox-Indicator Visual Baseline (Vietnamese)

This repository contains the visual-only baseline using the target-region bbox indicator input path.

Architecture

full image + target region bbox [cx, cy, w, h] / 1000
  -> SigLIP2 patch tokens
  -> Linear(4, 1024) bbox indicator token
  -> prepend bbox indicator before patch tokens
  -> semantic projector
  -> mBART decoder with LoRA
  -> Vietnamese region caption

Main files

model_state.pt
model_config.json
tokenizer/
image_processor/
bbox_indicator_baseline/
requirements.txt
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support