Papers › Breaking Common Sense: WHOOPS! A Vision-and-Language Benchmark of Synthetic and...

Breaking Common Sense: WHOOPS! A Vision-and-Language Benchmark of Synthetic and Compositional Images

13 Mar 2023ICCV 2023 1arXiv:2303.07274archive 2025-07-28

Nitzan Bitton-Guetta, Yonatan Bitton, Jack Hessel, Ludwig Schmidt, Yuval Elovici, Gabriel Stanovsky, Roy Schwartz

Weird, unusual, and uncanny images pique the curiosity of observers because they challenge commonsense. For example, an image released during the 2022 world cup depicts the famous soccer stars Lionel Messi and Cristiano Ronaldo playing chess, which playfully violates our expectation that their competition should occur on the football field. Humans can easily recognize and interpret these unconventional images, but can AI models do the same? We introduce WHOOPS!, a new dataset and benchmark for visual commonsense. The dataset is comprised of purposefully commonsense-defying images created by designers using publicly-available image generation tools like Midjourney. We consider several tasks posed over the dataset. In addition to image captioning, cross-modal matching, and visual question answering, we introduce a difficult explanation generation task, where models must identify and explain why a given image is unusual. Our results show that state-of-the-art models such as GPT3 and BLIP2 still lag behind human performance on WHOOPS!. We hope our dataset will inspire the development of AI models with stronger visual commonsense reasoning abilities. Data, models and code are available at the project website: whoops-benchmark.github.io

PaperPDFConference PDF

In Syntology View this paper on Syntology: its repositories, every harvested function with whether it ran, its licence and the call to fetch it.

Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

No code repository is listed for this paper in the archive or in Syntology's graph.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Common Sense ReasoningExplanation GenerationImage CaptioningImage GenerationImage-to-Text RetrievalQuestion AnsweringVisual Commonsense ReasoningVisual Question AnsweringVisual Question Answering (VQA)

Datasets

Introduced by this paper, per the archive.

WHOOPS!

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Explanation Generation WHOOPS! Ground-truth Caption -> GPT3 (Oracle) Human (%) 68 #1 of 7 Archive leaderboard report
Explanation Generation WHOOPS! Predicted Caption -> GPT3 Human (%) 33 #2 of 7 Archive leaderboard report
Explanation Generation WHOOPS! BLIP2 FlanT5-XXL (Fine-tuned) Human (%) 27 #3 of 7 Archive leaderboard report
Explanation Generation WHOOPS! BLIP2 FlanT5-XL (Fine-tuned) Human (%) 15 #4 of 7 Archive leaderboard report
Explanation Generation WHOOPS! BLIP2 FlanT5-XXL (Zero-shot) Human (%) 0 #5 of 7 Archive leaderboard report
Image Captioning WHOOPS! BLIP2 FlanT5-XXL (Fine-tuned) BLEU-4 42 #1 of 6 Archive leaderboard report
Image Captioning WHOOPS! BLIP2 FlanT5-XXL (Fine-tuned) CIDEr 177 #1 of 6 Archive leaderboard report
Image Captioning WHOOPS! BLIP2 FlanT5-XL (Fine-tuned) BLEU-4 41 #2 of 6 Archive leaderboard report
Image Captioning WHOOPS! BLIP2 FlanT5-XL (Fine-tuned) CIDEr 174 #2 of 6 Archive leaderboard report
Image Captioning WHOOPS! BLIP2 FlanT5-XXL (Zero-Shot) BLEU-4 31 #3 of 6 Archive leaderboard report
Image Captioning WHOOPS! BLIP2 FlanT5-XXL (Zero-Shot) CIDEr 120 #3 of 6 Archive leaderboard report
Image Captioning WHOOPS! CoCa ViT-L-14 MSCOCO BLEU-4 25 #4 of 6 Archive leaderboard report
Image Captioning WHOOPS! CoCa ViT-L-14 MSCOCO CIDEr 102 #4 of 6 Archive leaderboard report
Image Captioning WHOOPS! BLIP Large BLEU-4 13 #5 of 6 Archive leaderboard report
Image Captioning WHOOPS! BLIP Large CIDEr 65 #5 of 6 Archive leaderboard report
Image Captioning WHOOPS! OFA Large BLEU-4 0 #6 of 6 Archive leaderboard report
Image Captioning WHOOPS! OFA Large CIDEr 0 #6 of 6 Archive leaderboard report
Image-to-Text Retrieval WHOOPS! BLIP2 FlanT5-XXL (Text-only FT) Specificity 94 #1 of 7 Archive leaderboard report
Image-to-Text Retrieval WHOOPS! BLIP2 FlanT5-XXL (Fine-tuned) Specificity 84 #2 of 7 Archive leaderboard report
Image-to-Text Retrieval WHOOPS! BLIP2 FlanT5-XL (Fine-tuned) Specificity 81 #3 of 7 Archive leaderboard report
Image-to-Text Retrieval WHOOPS! BLIP Large Specificity 77 #4 of 7 Archive leaderboard report
Image-to-Text Retrieval WHOOPS! CoCa ViT-L-14 MSCOCO Specificity 72 #5 of 7 Archive leaderboard report
Image-to-Text Retrieval WHOOPS! BLIP2 FlanT5-XXL (Zero-shot) Specificity 71 #6 of 7 Archive leaderboard report
Image-to-Text Retrieval WHOOPS! CLIP ViT-L/14 Specificity 70 #7 of 7 Archive leaderboard report
Visual Question Answering (VQA) WHOOPS! BLIP2 FlanT5-XXL (Fine-tuned) BEM 57 #1 of 6 Archive leaderboard report
Visual Question Answering (VQA) WHOOPS! BLIP2 FlanT5-XXL (Fine-tuned) Exact Match 21 #1 of 6 Archive leaderboard report
Visual Question Answering (VQA) WHOOPS! BLIP2 FlanT5-XL (Fine-tuned) BEM 55 #2 of 6 Archive leaderboard report
Visual Question Answering (VQA) WHOOPS! BLIP2 FlanT5-XL (Fine-tuned) Exact Match 20 #2 of 6 Archive leaderboard report
Visual Question Answering (VQA) WHOOPS! BLIP2 FlanT5-XXL (Zero-shot) BEM 55 #3 of 6 Archive leaderboard report
Visual Question Answering (VQA) WHOOPS! BLIP2 FlanT5-XXL (Zero-shot) Exact Match 15 #3 of 6 Archive leaderboard report
Visual Question Answering (VQA) WHOOPS! OFA Large BEM 38 #4 of 6 Archive leaderboard report
Visual Question Answering (VQA) WHOOPS! OFA Large Exact Match 8 #4 of 6 Archive leaderboard report
Visual Question Answering (VQA) WHOOPS! BLIP Large BEM 39 #5 of 6 Archive leaderboard report
Visual Question Answering (VQA) WHOOPS! BLIP Large Exact Match 6 #5 of 6 Archive leaderboard report
Visual Question Answering (VQA) WHOOPS! BLIP2 FlanT5-XXL (Text-only FT) BEM 24 #6 of 6 Archive leaderboard report
Visual Question Answering (VQA) WHOOPS! BLIP2 FlanT5-XXL (Text-only FT) Exact Match 4 #6 of 6 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

1x1 ConvolutionAdamBatch NormalizationBigGANConditional Batch NormalizationConvolutionDense ConnectionsEarly StoppingFeedforward NetworkGAN Hinge LossLinear LayerNon-Local BlockNon-Local OperationOff-Diagonal Orthogonal RegularizationProjection DiscriminatorReLUResidual BlockResidual ConnectionSAGANSoftmaxSpectral NormalizationTTURTruncation Trick

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections