Browse State-of-the-Art › Image Captioning

Image Captioning

774 papers with code · 33 benchmarks · 79 datasets archive 2025-07-28

Computer VisionNatural Language Processing

Image Captioning is the task of describing the content of an image in words. This task lies at the intersection of computer vision and natural language processing. Most image captioning systems use an encoder-decoder framework, where an input image is encoded into an intermediate representation of the information in the image, and then decoded into a descriptive text sequence. The most popular benchmarks are nocaps and COCO, and models are typically evaluated according to a BLEU or CIDER metric.

( Image credit: Reflective Decoding Network for Image Captioning, ICCV'19)

Description from the archive archive 2025-07-28.

Benchmarks archive 2025-07-28

34 leaderboard tables shown for this task, 33 with rows (a “benchmark” on this site is a table with at least one row, as on /sota), ordered by row count. “Best model” is the first row in the archive's own order at snapshot; nothing is re-ranked here and metric direction is not recorded in the archive. PwC's Trend sparklines are not in the archive, so that column is omitted. 10 shown of 34 until expanded.

Syntology 7 newer papers were judged to report on this task's benchmarks so far (Claude Sonnet 4.5, reading their tables: a model's judgement, not a result; 0 placed): nocaps in-domain (3 judged, 0 placed), nocaps near-domain (3 judged, 0 placed), nocaps out-of-domain (3 judged, 0 placed), nocaps entire (3 judged, 0 placed), COCO (Common Objects in Context) (2 judged, 0 placed), nocaps-val-overall (1 judged, 0 placed), Flickr30k Captions test (3 judged, 0 placed). Drawn from 8,886 of 18,637 newer papers checked so far (733 more had nothing to place; 41 could not be read).

DatasetBest model (first row in archive order)PaperCodeSyntologyCompare
VizWiz 2020 test-dev (50 rows) IBM Research AI — — — Compare
COCO Captions (41 rows) mPLUG mPLUG: Effective and Efficient Vision-Language Learning by... code — Compare
nocaps in-domain (41 rows) PaLI PaLI: A Jointly-Scaled Multilingual Language-Image Model code Syntology official (archive's flag): 2 ran · 2 ran (of which 1 constructed an object rather than computing a result; 2 with no instrument failure: 0 honoured, 1 violated, 1 with no contract checked; 0 where Syntology's instrument failed) · 2 unverified Compare
nocaps near-domain (40 rows) GIT2, Single Model GIT: A Generative Image-to-text Transformer for Vision and Language code Syntology official (archive's flag): 14 ran · 14 ran (of which 0 constructed an object rather than computing a result; 14 with no instrument failure: 0 honoured, 0 violated, 14 with no contract checked; 0 where Syntology's instrument failed) · 7 unverified Compare
nocaps out-of-domain (40 rows) PaLI PaLI: A Jointly-Scaled Multilingual Language-Image Model code Syntology official (archive's flag): 2 ran · 2 ran (of which 1 constructed an object rather than computing a result; 2 with no instrument failure: 0 honoured, 1 violated, 1 with no contract checked; 0 where Syntology's instrument failed) · 2 unverified Compare
nocaps entire (39 rows) Lyrics Lyrics: Boosting Fine-grained Language-Vision Alignment and... — — Compare
COCO (Common Objects in Context) (17 rows) ExpansionNet v2 Exploiting Multiple Sequence Lengths in Fast End to End Training... code — Compare
VizWiz 2020 test (13 rows) IBM Research AI — — — Compare
nocaps-XD entire (12 rows) GIT2 GIT: A Generative Image-to-text Transformer for Vision and Language code Syntology official (archive's flag): 14 ran · 14 ran (of which 0 constructed an object rather than computing a result; 14 with no instrument failure: 0 honoured, 0 violated, 14 with no contract checked; 0 where Syntology's instrument failed) · 7 unverified Compare
nocaps-val-in-domain (11 rows) BLIP-2 ViT-G FlanT5 XL (zero-shot) BLIP-2: Bootstrapping Language-Image Pre-training with Frozen... code Syntology community repositories only · 4 ran (of which 4 constructed an object rather than computing a result; 4 with no instrument failure: 0 honoured, 0 violated, 4 with no contract checked; 0 where Syntology's instrument failed) · 4 unverified; every one of the 4 samples that ran constructed an object rather than computing a result Compare
nocaps-val-overall (11 rows) BLIP-2 ViT-G FlanT5 XL (zero-shot) BLIP-2: Bootstrapping Language-Image Pre-training with Frozen... code Syntology community repositories only · 4 ran (of which 4 constructed an object rather than computing a result; 4 with no instrument failure: 0 honoured, 0 violated, 4 with no contract checked; 0 where Syntology's instrument failed) · 4 unverified; every one of the 4 samples that ran constructed an object rather than computing a result Compare
nocaps-XD in-domain (11 rows) GIT2 GIT: A Generative Image-to-text Transformer for Vision and Language code Syntology official (archive's flag): 14 ran · 14 ran (of which 0 constructed an object rather than computing a result; 14 with no instrument failure: 0 honoured, 0 violated, 14 with no contract checked; 0 where Syntology's instrument failed) · 7 unverified Compare
nocaps-XD near-domain (11 rows) GIT2 GIT: A Generative Image-to-text Transformer for Vision and Language code Syntology official (archive's flag): 14 ran · 14 ran (of which 0 constructed an object rather than computing a result; 14 with no instrument failure: 0 honoured, 0 violated, 14 with no contract checked; 0 where Syntology's instrument failed) · 7 unverified Compare
nocaps-XD out-of-domain (11 rows) GIT2 GIT: A Generative Image-to-text Transformer for Vision and Language code Syntology official (archive's flag): 14 ran · 14 ran (of which 0 constructed an object rather than computing a result; 14 with no instrument failure: 0 honoured, 0 violated, 14 with no contract checked; 0 where Syntology's instrument failed) · 7 unverified Compare
TextCaps 2020 (11 rows) TAP — — — Compare
nocaps-val-near-domain (10 rows) BLIP-2 ViT-G FlanT5 XL (zero-shot) BLIP-2: Bootstrapping Language-Image Pre-training with Frozen... code Syntology community repositories only · 4 ran (of which 4 constructed an object rather than computing a result; 4 with no instrument failure: 0 honoured, 0 violated, 4 with no contract checked; 0 where Syntology's instrument failed) · 4 unverified; every one of the 4 samples that ran constructed an object rather than computing a result Compare
nocaps-val-out-domain (10 rows) BLIP-2 ViT-G FlanT5 XL (zero-shot) BLIP-2: Bootstrapping Language-Image Pre-training with Frozen... code Syntology community repositories only · 4 ran (of which 4 constructed an object rather than computing a result; 4 with no instrument failure: 0 honoured, 0 violated, 4 with no contract checked; 0 where Syntology's instrument failed) · 4 unverified; every one of the 4 samples that ran constructed an object rather than computing a result Compare
SCICAP (9 rows) CNN+LSTM (Vision only, First sentence) SciCap: Generating Captions for Scientific Figures code — Compare
Flickr30k Captions test (7 rows) Unified VLP Unified Vision-Language Pre-Training for Image Captioning and VQA code Syntology official (archive's flag): 11 ran · 11 ran (of which 0 constructed an object rather than computing a result; 6 with no instrument failure: 1 honoured, 0 violated, 5 with no contract checked; 5 where Syntology's instrument failed) · 3 unverified Compare
WHOOPS! (6 rows) BLIP2 FlanT5-XXL (Fine-tuned) Breaking Common Sense: WHOOPS! A Vision-and-Language Benchmark of... — — Compare
nocaps val (3 rows) Prismer Prismer: A Vision-Language Model with Multi-Task Experts code — Compare
Object HalBench (3 rows) RLHF-V RLHF-V: Towards Trustworthy MLLMs via Behavior Alignment from... code — Compare
COCO Captions test (2 rows) From Captions to Visual Concepts and Back From Captions to Visual Concepts and Back code Syntology official (archive's flag): 2 ran · 2 ran (of which 0 constructed an object rather than computing a result; 2 with no instrument failure: 0 honoured, 0 violated, 2 with no contract checked; 0 where Syntology's instrument failed) · 4 unverified Compare
Conceptual Captions (2 rows) ClipCap (MLP + GPT2 tuning) ClipCap: CLIP Prefix for Image Captioning code Syntology official (archive's flag): 5 ran · 6 ran (of which 0 constructed an object rather than computing a result; 5 with no instrument failure: 0 honoured, 0 violated, 5 with no contract checked; 1 where Syntology's instrument failed) · 0 unverified Compare
FlickrStyle10K (2 rows) CapDec Text-Only Training for Image Captioning using Noise-Injected CLIP code Syntology official (archive's flag): 2 ran · 4 ran (of which 0 constructed an object rather than computing a result; 2 with no instrument failure: 1 honoured, 1 violated, 0 with no contract checked; 2 where Syntology's instrument failed) · 1 unverified Compare
Localized Narratives (2 rows) LoopCAG Control Image Captioning Spatially and Temporally — — Compare
AIC-ICC (1 row) CMCL WenLan: Bridging Vision and Language by Large-Scale Multi-Modal... code — Compare
BanglaLekhaImageCaptions (1 row) CNN + 1D CNN Improved Bengali Image Captioning via deep convolutional neural... code — Compare
ChEBI-20 (1 row) GIT-Mol GIT-Mol: A Multi-modal Large Language Model for Molecular Science... code Syntology official (archive's flag): 1 ran · 1 ran (of which 0 constructed an object rather than computing a result; 0 with no instrument failure: 0 honoured, 0 violated, 0 with no contract checked; 1 where Syntology's instrument failed) · 0 unverified Compare
IU X-Ray (1 row) BiomedGPT BiomedGPT: A Generalist Vision-Language Foundation Model for... code Syntology official (archive's flag): 6 ran · 6 ran (of which 0 constructed an object rather than computing a result; 6 with no instrument failure: 1 honoured, 0 violated, 5 with no contract checked; 0 where Syntology's instrument failed) · 2 unverified Compare
MS-COCO (1 row) NeuSyRE NeuSyRE: Neuro-Symbolic Visual Understanding and Reasoning... code — Compare
MSCOCO (1 row) CapDec Text-Only Training for Image Captioning using Noise-Injected CLIP code Syntology official (archive's flag): 2 ran · 4 ran (of which 0 constructed an object rather than computing a result; 2 with no instrument failure: 1 honoured, 1 violated, 0 with no contract checked; 2 where Syntology's instrument failed) · 1 unverified Compare
Peir Gross (1 row) BiomedGPT BiomedGPT: A Generalist Vision-Language Foundation Model for... code Syntology official (archive's flag): 6 ran · 6 ran (of which 0 constructed an object rather than computing a result; 6 with no instrument failure: 1 honoured, 0 violated, 5 with no contract checked; 0 where Syntology's instrument failed) · 2 unverified Compare
foundation-multimodal-models/DetailCaps-4870 (0 rows) no rows in the archive — —

Syntology column: samples harvested from the paper's repositories and executed on synthesized fixtures; “ran” is not a correctness claim and does not order the table. Each line reads “N ran (of which C constructed an object rather than computing a result; K with no instrument failure: H honoured, V violated, P with no contract checked; I where Syntology's instrument failed) · U unverified”; the instrument figure counts failures of Syntology's instrument, not of the code. When the archive marks a repository official for the row's paper, the line starts with that repository's state (“official” is the archive's flag, not a verdict on who wrote the code; “community repositories only” when every sample that ran came from a community repository, “official: no sample here; runs from other or unrecorded repositories” when some came from a repository the paper names or has in its text, or from none recorded); hover it for the repositories the samples that ran came from. A dash means no Syntology record for that paper, not a recorded non-run. Each “ran” line links to that paper's sample list. Syntology's record for this page last changed ; when this build read Syntology's graph is in the build record. For agents: get_harvested_code_for_paper(arxiv_id) lists each paper's samples; how to connect.

Libraries

Not in the archive: the export carries no per-task library table, so there is nothing to show at snapshot 2025-07-28.

Datasets archive 2025-07-28

79 datasets whose archive record lists this task, ordered by the archive's paper count. 30 shown of 79 until expanded.

Subtasks archive 2025-07-28

8 subtasks in the archive's task tree.

Papers archive 2025-07-28

archive papers tagged: 1,878 · with a code link: 774 · where Syntology ran a sample: 243 (194 with a run with no instrument failure, 49 where every run was a failure of Syntology's instrument) Syntology

Show: all tagged papersonly where code ran (243 of 1,878 tagged: 194 with a run with no instrument failure, 49 where every run was a failure of Syntology's instrument)

Page 1 of 19: papers 1 to 100 of 1,878, in archive order: by repositories listed in the archive (most first), then newest first, not by stars (the archive holds no stars, so PwC's “Social” and “Latest” sorts cannot be reproduced). Papers that list no repository come after every paper that lists one.

Papers without a page here are shown as plain text. A Syntology line reads “N ran (of which C constructed an object rather than computing a result; K with no instrument failure: H honoured, V violated, P with no contract checked; I where Syntology's instrument failed) · U unverified”; the figure “where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. When the archive marks a repository official for the paper, the line starts with that repository's state (the archive's flag, not a verdict on who wrote the code); hover it for the repositories the samples that ran came from. The first sentence of the abstract is shown for the first 30 papers; every abstract is on its paper's page.

Syntology lines on 64 of the papers shown; no Syntology record for the others (a paper without an arXiv id cannot be joined to the graph, and absence from the graph layer is not a recorded non-run). “Ran” means the sample executed on a synthesized fixture, not that the paper's result was reproduced; each line links to that paper's sample list. Syntology's record for this page last changed ; when this build read Syntology's graph is in the build record. For agents: get_harvested_code_for_paper(arxiv_id) lists each paper's samples; how to connect.

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