Research map / Vision-language and multimodal / Multimodal LLMs
The Abduction of Sherlock Holmes: A Dataset for Visual Abductive Reasoning
Jack Hessel, Jena D. Hwang, Jae Sung Park, Rowan Zellers, Chandra Bhagavatula, Anna Rohrbach, Kate Saenko, Yejin Choi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add multitask clue learning | CLIP RN50x64 · Sherlock retrieval | P@1 (im->txt): 31.8 -> 33.4 | Improved |
| add multitask clue learning | CLIP RN50x64 · Sherlock comparison (val) | human accuracy: 25.1 -> 26.6 | Improved |
| scale CLIP backbone from ViT-B/16 to RN50x64 | CLIP · Sherlock comparison (val) | human accuracy: 20.1 -> 25.1 | Improved |
| use CLIP instead of UNITER Base | CLIP ViT-B/16 · Sherlock retrieval | P@1 (im->txt): 19.8 -> 30.6 | Improved |
| pass the region as input (vs not passing region) (ablation) | CLIP ViT-B/16 · Sherlock retrieval | P@1 (im->txt): 18 -> 31 | Improved |
| keep surrounding context rather than only the region (No Context ablation) (ablation) | CLIP ViT-B/16 · Sherlock | retrieval and comparison | Improved |
| keep region content rather than only context (Only Context ablation) (ablation) | CLIP ViT-B/16 · Sherlock | retrieval and comparison | Improved |
| train on inferences rather than clues only (ablation) | CLIP ViT-B/16 · Sherlock | retrieval and comparison | Improved |
| use two-pass widescreen processing instead of center cropping or pad-and-resize (ablation) | CLIP ViT-B/16 · Sherlock retrieval | P@1 | Improved |
| use two-pass widescreen processing instead of center cropping or pad-and-resize (ablation) | CLIP ViT-B/16 · Sherlock comparison | human accuracy | No clear effect |
| train on clues rather than inferences, test on clues | CLIP RN50x64 · Sherlock clue retrieval | P@1: 43.4 -> 54.3 | Improved |
| add clue-to-inference text-only oracle versus image+text model | T5-Large oracle · Sherlock comparison (val) | human accuracy: 26.6 -> 22.8 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- UNITER Base — UNITER: UNiversal Image-TExt Representation Learning (ECCV 2020) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- AbductiveMLLM: Boosting Visual Abductive Reasoning Within MLLMs (AAAI 2026 · similarity 0.92)
- NL-Eye: Abductive NLI For Images (ICLR 2025 · similarity 0.89)
- HumanPCR: Probing MLLM Capabilities in Diverse Human-Centric Scenes (ICLR 2026 · similarity 0.88)
- Black Swan: Abductive and Defeasible Video Reasoning in Unpredictable Events (CVPR 2025 · similarity 0.88)
- Mind the GAP: Glimpse-based Active Perception improves generalization and sample efficiency of visual reasoning (ICLR 2025 · similarity 0.87)
- Machine Mental Imagery: Empower Multimodal Reasoning with Latent Visual Tokens (CVPR 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
