Papers › DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning

DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning

20 May 2025arXiv:2505.14362archive 2025-07-28

Ziwei Zheng, Michael Yang, Jack Hong, Chenxiao Zhao, Guohai Xu, Le Yang, Chao Shen, Xing Yu

Large Vision-Language Models (VLMs) have shown strong capabilities in multimodal understanding and reasoning, yet they are primarily constrained by text-based reasoning processes. However, achieving seamless integration of visual and textual reasoning which mirrors human cognitive processes remains a significant challenge. In particular, effectively incorporating advanced visual input processing into reasoning mechanisms is still an open question. Thus, in this paper, we explore the interleaved multimodal reasoning paradigm and introduce DeepEyes, a model with "thinking with images" capabilities incentivized through end-to-end reinforcement learning without the need for cold-start SFT. Notably, this ability emerges natively within the model itself, leveraging its inherent grounding ability as a tool instead of depending on separate specialized models. Specifically, we propose a tool-use-oriented data selection mechanism and a reward strategy to encourage successful tool-assisted reasoning trajectories. DeepEyes achieves significant performance gains on fine-grained perception and reasoning benchmarks and also demonstrates improvement in grounding, hallucination, and mathematical reasoning tasks. Interestingly, we observe the distinct evolution of tool-calling behavior from initial exploration to efficient and accurate exploitation, and diverse thinking patterns that closely mirror human visual reasoning processes. Code is available at https://github.com/Visual-Agent/DeepEyes.

PaperPDFCodeCode Syntology ran

In Syntology View this paper on Syntology, its page in Syntology's graph. That page adds which of its functions are also present in other papers' code Syntology holds, with dates, and its dated run record as JSON and a README badge. It also lists the repositories linked to the paper, the abstract and the calls for agents.

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

For agents, on Syntology's MCP service (how to connect):

Code

Syntology 17 code samples harvested from 1 repository linked to this paper: 12 ran (of which 0 constructed an object rather than computing a result; 8 with no instrument failure: 1 honoured, 0 violated, 7 with no contract checked; 4 where Syntology's instrument failed) · 5 unverified. Where Syntology's instrument failed: 4 where our contract draft was wrong; failures of Syntology's instrument, not of the code.

By repository: official repository: 17 samples from 1 repository, 12 ran (8 with no instrument failure; 4 where Syntology's instrument failed). The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

Levels written between 2026-09-22 and 2026-09-25 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed.

visual-agent/deepeyes officialmentioned in papermentioned on GitHubpytorchApache-2.0Syntology: 17 samples, 12 ran (4 instrument-failed) · 5 unverified report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

17 samples harvested: 12 ran (of which 0 constructed an object rather than computing a result; 8 with no instrument failure: 1 honoured, 0 violated, 7 with no contract checked; 4 where Syntology's instrument failed) · 5 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Syntology's record for this page last changed ; when this build read Syntology's graph is in the build record. That date is not when the samples ran.

1ran · honoured contract
4ran · our draft was wrong
7ran
5unverified

Licence: 4 of the 17 samples are pointer only, meaning Syntology does not serve that copy's text. For 4 of them, this paper's copy is under a permissive licence and cleared, but the stored copy of the same code that Syntology serves is not; the licence label's tooltip gives Syntology's recorded reason. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from visual-agent/deepeyes. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. A line beginning “fixture:” under a sample that ran gives the arguments Syntology's fixture called the sample with, each as name:type/rank/element kind (? where none was recorded), and the type of what the fixture's run returned, exactly as the harness recorded them; the arguments and the returned type were recorded by two separate runs of the same fixture. No output value is shown, and a returned type says nothing about whether the output was right. “Level recorded” gives the UTC date when Syntology's graph wrote the sample's verification level, not when the code executed; where the level was written before Syntology's graph kept that date, the row says “level recorded, date not recorded”. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself. “Our draft was wrong” and “the contract check could not run” are failures of Syntology's instrument, not of the code: the sample ran on Syntology's synthesized input, and then the drafted contract did not describe it, or the check itself raised an error (RAISES), lacked a dependency (DEP_MISSING) or ran out of time (TIMEOUT); hover the label for which. A metamorphic tier, where shown, is evidence of the output's shape, not that the sample implements the method.

Each sample ends with its code_sha256. A sample's code_sha256 is Syntology's identity for that exact code. It is printed in full (16 hexadecimal characters for every sample in this build), and get_code(code_sha256="…") takes it exactly as printed. An agent calls it on Syntology's MCP service (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted. community: Not in the archive's code links for this paper; a community repository Syntology harvested. Samples are grouped by repository: a repository marked official first, then by how directly the repository is tied to the paper, then in the archive's order; never by how many ran. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of six reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar); this paper's copy was not located, so its licence is not recorded; or this paper's copy is cleared, but the stored copy of the same code that Syntology serves is not. Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

A function or class name links to its page when the name occurs in the harvested code of at least 5 papers; the name is a string, not an identity.

visual-agent/deepeyes official repository 17 samples, 12 ran (4 instrument-failed) · 5 unverified
compute_ce_dpo_loss_rm in 48 papers recipe/prime/prime_core_algos.py ran Apache-2.0 (permissive) here; the stored copy is pointer only · f220595adfc5e83c · report fixture: signature token_level_scores:Tensor/2/float, acc:Tensor/1/float, response_mask:Tensor/2/float, beta:float/0/float · returned Tensor · level recorded 2026-09-24
compute_detach_dpo_loss_rm in 49 papers recipe/prime/prime_core_algos.py ran · our draft was wrong Apache-2.0 (permissive) here; the stored copy is pointer only · 4d91e4383716e42c · report fixture: signature token_level_scores:Tensor/2/float, acc:Tensor/1/float, Q_bc:list/1/Tensor, acc_bc:Tensor/2/float, response_mask:Tensor/2/float, beta:float/0/float, bon_mode:str/0/str · returned Tensor · level recorded 2026-09-24
decode_base64_to_image in 8 papers eval/eval_hrbench.py ran Apache-2.0 (permissive) · 42fa1c710986d2ee · report fixture: signature base64_string:str/0/str, target_size:int/0/int · returned PngImageFile · level recorded 2026-09-25
encode_image_to_base64 in 15 papers eval/eval_hrbench.py ran · our draft was wrong Apache-2.0 (permissive) · ae6c2bb194221137 · report fixture: signature image_path:str/0/str · returned str · level recorded 2026-09-25
encode_pil_image_to_base64 eval/eval_hrbench.py ran Apache-2.0 (permissive) · 988cae1e62ee4e6e · report fixture: signature pil_image:Image/0/? · returned str · level recorded 2026-09-25
encode_pil_image_to_base64 eval/eval_vstar.py ran Apache-2.0 (permissive) · 69e47c426e8167f5 · report fixture: signature pil_image:Image/0/? · returned str · level recorded 2026-09-25
get_custom_reward_fn in 13 papers recipe/dapo/src/main_dapo.py ran Apache-2.0 (permissive) · b6512040439cb7a2 · report fixture: signature config:dict/0/? · returned function · level recorded 2026-09-25
get_prompt in 99 papers eval/judge_result.py ran fingerprinted Apache-2.0 (permissive) · 1b4e5e1a7c7a8eee · report fixture: signature predict_str:str/0/str, ground_truth:str/0/str, question:str/0/str · returned str · level recorded 2026-09-25
reward_func in 49 papers recipe/r1/reward_score.py ran Apache-2.0 (permissive) · c30b2a96103abd2c · report fixture: signature data_source:str/0/str, solution_str:str/0/str, ground_truth:str/0/str, extra_info:NoneType/0/? · returned MagicMock · level recorded 2026-09-24
round_by_factor in 26 papers eval/eval_vstar.py ran · honoured contract fingerprinted Apache-2.0 (permissive) · e252767324188623 · report fixture: signature number:int/0/int, factor:int/0/int · returned int · level recorded 2026-09-22
union_tensor_dict in 69 papers verl/protocol.py ran · our draft was wrong Apache-2.0 (permissive) · 21331a58f93375e4 · report fixture: signature tensor_dict1:TensorDict/1/float, tensor_dict2:TensorDict/1/float · returned TensorDict · level recorded 2026-09-22
unpad_dataproto in 102 papers verl/protocol.py ran · our draft was wrong Apache-2.0 (permissive) · 25f0ea3f460f6ce1 · report fixture: signature data:DataProto/0/?, pad_size:int/0/int · returned DataProto · level recorded 2026-09-22
get_generation_config in 66 papers verl/utils/model.py unverified Apache-2.0 (permissive) here; the stored copy is pointer only · 68aa34a57b353cae · report
get_huggingface_actor_config in 83 papers verl/utils/model.py unverified Apache-2.0 (permissive) here; the stored copy is pointer only · 9411d1fb7e4f68e5 · report
get_weight_loader in 93 papers verl/models/weight_loader_registry.py unverified Apache-2.0 (permissive) · acf240079d434746 · report
get_weight_saver in 69 papers verl/models/weight_loader_registry.py unverified Apache-2.0 (permissive) · 1008c22d29f85015 · report
squeeze in 92 papers verl/utils/model.py unverified Apache-2.0 (permissive) · 3b15e2ac7497c441 · report

Tasks

HallucinationMathematical ReasoningMultimodal ReasoningReinforcement LearningVisual Reasoningreinforcement-learning

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

SFT

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