🖼️ Image Faithfulness Evaluator

Tab 1: Single prompt evaluation. Tab 2: Multi-image self-consistency. Tab 3: POPE Benchmark — binary hallucination labels. Tab 4: OK-VQA Benchmark — answer agreement on real images. Tab 5: Large-scale Flickr30k benchmark (live or CSV load).

📌 Tip for larger plots: Click the ⤢ expand icon in the top-right corner of any plot image to view it fullscreen. Use the ⬇ download button to save a high-resolution PNG for use in presentations.