REVERSE: Multi-Turn Tool-Augmented Reasoning for Image Geolocation

REVERSE is a dataset for training and evaluating vision-language models on image geolocation using multi-turn tool-augmented reasoning. It contains approximately 10,000 training images and over 2,200 supervised fine-tuning samples. The dataset was created by Jack11211 and was last updated on Hugging Face in May 2026.

Use Cases

• Train vision-language models on multi-turn reasoning for image geolocation based on the described task structure.
• Evaluate agentic models using simulated tools like zoom and search based on the dataset's purpose.
• Benchmark model performance on coldstart geolocation tasks using the provided supervised fine-tuning data.

Strengths

• Contains approximately 10,000 training images.
• Includes over 2,200 samples for supervised fine-tuning coldstart training.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Jack11211 on Hugging Face.

Freshness

Last updated 2026-05-15 10:12:54; freshness should be verified.