referring-expression
MRSeg-Referring-Expressions
MRSeg Referring Expressions (single-turn)
Single-turn referring-expression segmentation samples derived from the multi-round MR-Seg data
of SegLLM (paper).
SegLLM's original conversations provide the target regions as input to later rounds, which
makes them unusable for models that receive only an image and text. We used
Qwen3-VL-32B-Instruct to keep only the
samples whose target can be identified from the image and expression alone, rewriting each into
a self-contained… See the full description on the dataset page: https://huggingface.co/datasets/Panorama-grounding/MRSeg-Referring-Expressions.filtered_deepseek_v31_referring_expression_parsing
DeepSeek v3.1 Quality-Filtered Referring Expression Parsing + Distractor Labels
This dataset contains parsed referring expressions from the RefCOCO, RefCOCOg, and RefCOCO+ validation sets, processed using DeepSeek v3.1 with quality filtering, plus corresponding distractor label annotations in COCO format.
Dataset Description
Overview
Model: DeepSeek v3.1 (deepseek-chat)
Processing: Quality-filtered results from referring expression parsing
Datasets: RefCOCO… See the full description on the dataset page: https://huggingface.co/datasets/dddraxxx/filtered_deepseek_v31_referring_expression_parsing.