{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/tarsier-recipes-for-training-and-evaluating-1","title":"Tarsier: Recipes for Training and Evaluating Large Video Description Models","arxiv_id":"2407.00634","date":"2024-06-30","proceeding":"arXiv 2024 7","authors":["Jiawei Wang","Liping Yuan","Yuchen Zhang","Haomiao Sun"],"abstract":"Generating fine-grained video descriptions is a fundamental challenge in video understanding. In this work, we introduce Tarsier, a family of large-scale video-language models designed to generate high-quality video descriptions. Tarsier employs CLIP-ViT to encode frames separately and then uses an LLM to model temporal relationships. Despite its simple architecture, we demonstrate that with a meticulously designed two-stage training procedure, the Tarsier models exhibit substantially stronger video description capabilities than any existing open-source model, showing a $+51.4\\%$ advantage in human side-by-side evaluation over the strongest model. Additionally, they are comparable to state-of-the-art proprietary models, with a $+12.3\\%$ advantage against GPT-4V and a $-6.7\\%$ disadvantage against Gemini 1.5 Pro. When upgraded to Tarsier2 by building upon SigLIP and Qwen2-7B, it further improves significantly with a $+4.8\\%$ advantage against GPT-4o. Besides video description, Tarsier proves to be a versatile generalist model, achieving new state-of-the-art results across nine public benchmarks, including multi-choice VQA, open-ended VQA, and zero-shot video captioning. Our second contribution is the introduction of a new benchmark -- DREAM-1K (https://tarsier-vlm.github.io/) for evaluating video description models, consisting of a new challenging dataset featuring videos from diverse sources and varying complexity, along with an automatic method specifically designed to assess the quality of fine-grained video descriptions. We make our models and evaluation benchmark publicly available at https://github.com/bytedance/tarsier.","url_abs":"https://arxiv.org/abs/2407.00634v2","url_pdf":"https://arxiv.org/pdf/2407.00634v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"tarsier-recipes-for-training-and-evaluating-1","repo_url":"https://github.com/bytedance/tarsier","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"video-captioning","task_name":"Video Captioning"},{"task_slug":"video-description","task_name":"Video Description"},{"task_slug":"video-question-answering","task_name":"Video Question Answering"},{"task_slug":"video-understanding","task_name":"Video Understanding"},{"task_slug":"visual-question-answering","task_name":"Visual Question Answering (VQA)"},{"task_slug":"zeroshot-video-question-answer","task_name":"Zero-Shot Video Question Answer"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/video-question-answering-on-mvbench","task":"Video Question Answering","dataset":"MVBench","model":"Tarsier (34B)","rank_in_archive_order":2,"of":22,"metrics":{"Avg.":"67.6"},"uses_additional_data":false},{"leaderboard":"/sota/video-question-answering-on-tvbench","task":"Video Question Answering","dataset":"TVBench","model":"Tarsier-34B","rank_in_archive_order":7,"of":28,"metrics":{"Average Accuracy":"55.5"},"uses_additional_data":false},{"leaderboard":"/sota/video-question-answering-on-tvbench","task":"Video Question Answering","dataset":"TVBench","model":"Tarsier-7B","rank_in_archive_order":16,"of":28,"metrics":{"Average Accuracy":"46.9"},"uses_additional_data":false},{"leaderboard":"/sota/zeroshot-video-question-answer-on-activitynet","task":"Zero-Shot Video Question Answer","dataset":"ActivityNet-QA","model":"Tarsier (34B)","rank_in_archive_order":1,"of":28,"metrics":{"Accuracy":"61.6","Confidence Score":"3.7"},"uses_additional_data":false},{"leaderboard":"/sota/zero-shot-video-question-answer-on-egoschema-1","task":"Zero-Shot Video Question Answer","dataset":"EgoSchema (fullset)","model":"Tarsier (34B)","rank_in_archive_order":7,"of":29,"metrics":{"Accuracy":"61.7"},"uses_additional_data":false},{"leaderboard":"/sota/zero-shot-video-question-answer-on-egoschema","task":"Zero-Shot Video Question Answer","dataset":"EgoSchema (subset)","model":"Tarsier (34B)","rank_in_archive_order":1,"of":14,"metrics":{"Accuracy":"68.6"},"uses_additional_data":false},{"leaderboard":"/sota/zeroshot-video-question-answer-on-msrvtt-qa","task":"Zero-Shot Video Question Answer","dataset":"MSRVTT-QA","model":"Tarsier (34B)","rank_in_archive_order":5,"of":30,"metrics":{"Accuracy":"66.4","Confidence Score":"3.7"},"uses_additional_data":false},{"leaderboard":"/sota/zeroshot-video-question-answer-on-msvd-qa","task":"Zero-Shot Video Question Answer","dataset":"MSVD-QA","model":"Tarsier (34B)","rank_in_archive_order":1,"of":28,"metrics":{"Accuracy":"80.3","Confidence Score":"4.2"},"uses_additional_data":false},{"leaderboard":"/sota/zero-shot-video-question-answer-on-next-qa","task":"Zero-Shot Video Question Answer","dataset":"NExT-QA","model":"Tarsier (34B)","rank_in_archive_order":2,"of":27,"metrics":{"Accuracy":"79.2"},"uses_additional_data":false},{"leaderboard":"/sota/zeroshot-video-question-answer-on-tgif-qa","task":"Zero-Shot Video Question Answer","dataset":"TGIF-QA","model":"Tarsier (34B)","rank_in_archive_order":1,"of":14,"metrics":{"Accuracy":"82.5","Confidence Score":"4.4"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2407.00634","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2407.00634"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/bytedance/tarsier","reach":null}],"summary":{"ran_fixture":1,"ran_draft_wrong":1},"by_repo_kind":{"official":{"samples":2,"ran":2,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"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"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"ad08899b0ff95e16","entry":"add_split_tokens","repo":"bytedance/tarsier","repo_kind":"official","path":"models/modeling_tarsier.py","file_url":"https://github.com/bytedance/tarsier/blob/HEAD/models/modeling_tarsier.py","link_basis":"first_harvest_node","language":"python","status":"ran_fixture","verification_level":1,"contract_check":"RAISES","metamorphic_tier":"invariant","behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"ad08899b0ff95e16"}},{"code_sha256_prefix":"44a7aa16432213cb","entry":"extract_item_for_eval","repo":"bytedance/tarsier","repo_kind":"official","path":"evaluation/evaluate.py","file_url":"https://github.com/bytedance/tarsier/blob/HEAD/evaluation/evaluate.py","link_basis":"first_harvest_node","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"44a7aa16432213cb"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}