EvOps Papers

Research question: LMS/VLMを共通評価器として使い、評価結果をDBへ蓄積し、MLによる改善へ接続できるか。

カードを大きく表示。クライテリアを選ぶと、同じ評価軸の論文がまとまって並びます。

External Papers

P01 — VeMo: Zero-Shot Text-to-Motion Evaluation using Video Language Models

Text-to-motion と生成動画の意味一致を VLM で評価

ICML 2026 / 2026 / PDF unavailable

P02 — EvalCrafter: Benchmarking and Evaluating Large Video Generation Models

動画生成モデルを多軸で評価するベンチマーク

CVPR 2024 / 2024 / PDF unavailable

P03 — VideoScore: Building Automatic Metrics to Simulate Fine-grained Human Feedback for Video Generation

細粒度な人間評価を自動評価へ変換

EMNLP 2024 / 2024 / PDF unavailable

P04 — VisualPrompter

視覚的評価から不足概念を発見・プロンプト改善へ接続

ICLR 2026 / 2026 / PDF unavailable

P05 — 3D Human Animation Quality Assessment: Subjective and Objective Evaluation

3D アニメーションの主観・客観評価と品質予測

IEEE TVCG 2026 / 2026 / PDF unavailable

P06 — DreamFusion: Text-to-3D using 2D Diffusion

Score Distillation Sampling でテキストから 3D 造型を最適化

ICCV 2023 / 2023 / PDF unavailable

P07 — Point-E: A System for Generating 3D Point Clouds from Textual Descriptions

Text-to-3D の高速生成パイプライン

CVPR 2023 / 2023 / PDF unavailable

P08 — Magic3D: High-Resolution Text-to-3D Content Creation

多段階生成→upscale で高解像度化

CVPR 2024 / 2024 / PDF unavailable

P09 — What matters when building vision-language models? (Idefics2)

VLM 設計の決定要因を実験的に明らかに

arXiv 2024 / 2024 / PDF unavailable

P10 — CLIP: Learning Transferable Visual Models From Natural Language Supervision

画像とテキストを共通空間に埋め込むゼロショット評価の基盤

ICML 2021 / 2021 / PDF unavailable

P11 — Pick-a-Pic: An Open Dataset of User Preferences for Text-to-Image Generation

テキスト画像生成のための大規模嗜好データセット

NeurIPS 2023 / 2023 / PDF unavailable

P12 — ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation

人間の嗜好を学習したテキスト画像評価モデル

NeurIPS 2023 / 2023 / PDF unavailable

P13 — Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesis

テキスト画像生成の人間嗜好ベンチマーク

arXiv 2023 / 2023 / PDF unavailable

P14 — Training language models to follow instructions with human feedback (InstructGPT / RLHF)

人間フィードバックからの強化学習による指示追従

NeurIPS 2022 / 2022 / PDF unavailable

P15 — Direct Preference Optimization: Your Language Model is Secretly a Reward Model

報酬モデルなしで嗜好データから直接最適化

NeurIPS 2024 / 2024 / PDF unavailable

P16 — Active Learning for Large Language Model-based Chatbots

LLM チャットボットのためのアクティブラーニング

EMNLP 2023 / 2023 / PDF unavailable

P17 — Model Context Protocol (MCP)

LLM と外部ツール・データソースを標準接続するプロトコル

Anthropic Technical Specification / 2024 / PDF unavailable

P18 — Agent2Agent Protocol (A2A)

エージェント間の相互運用プロトコル

Google Technical Specification / 2025 / PDF unavailable

P19 — Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone

小規模・高品質な言語モデル

arXiv 2024 / 2024 / PDF unavailable

P20 — Llama 3.2: Revolutionizing edge AI and vision with open, Customizable Models

エッジデバイス向け軽量マルチモーダルモデル

Meta Technical Report / 2024 / PDF unavailable

P21 — The False Dawn: Reevaluating Performance Gains of Small Language Models

小規模言語モデルの実用上の限界と可能性

arXiv 2024 / 2024 / PDF unavailable

P22 — A Watermark for Large Language Models

LLM 出力の追跡可能な透かし

ICML 2023 / 2023 / PDF unavailable

P25 — B-CODER: Benchmarking Code Generation with Evaluation Loop

評価ループ付きコード生成ベンチマーク

ICLR 2024 / 2024 / PDF unavailable

EvOps Original Papers

P23 — VQualA: Video Quality Assessment via VLM

VLMを用いた動画品質評価フレームワーク

in-house / 2025 / Doc

P24 — UniAPO: Unified Automatic Pipeline Orchestration

統合自動パイプラインオーケストレーション

in-house / 2025 / Doc

P26 — MLOps Evaluation Loop Automation

MLOps評価ループの自動化設計

in-house / 2024 / Doc

P27 — VLM Evaluation Automation for Visual Content

視覚コンテンツ向けVLM評価自動化

in-house / 2024 / Doc

P28 — Evolutionary Generation Loop with Feedback Accumulation

フィードバック蓄積型進化生成ループ

in-house / 2024 / Doc

P29 — Jev-Driven MLOps: High-Speed Judgment Layer for Observable and Controllable Generative Pipelines

MLOps内部への高速判断レイヤー統合

in-house / 2024 / Doc

P30 — Model-Free Visual Synthesis: Deterministic 3D Rendering via ComfyUI-Workflow Concepts Mapped to Blender-bpy

拡散モデル不要の決定論的ビジュアル合成

in-house / 2024 / Doc

P31 — Inverse Preference Learning for Evaluation Agents: Inferring Human Value Functions from Scores

人間の価値関数を採点履歴から逆推定

in-house / 2024 / Doc

P32 — Cost-Aware Rotation Loop Design for Resource-Constrained MLOps: Token Budget Optimization in CPU-Only Generative Pipelines

CPU制約下でのコスト意識型回転ループ設計

in-house / 2024 / Doc

P33 — Structured Feature Axes for Character-Aware Visual Correction Loops: Multi-Layer Evaluation Combining Detection, Histogram, and VLM

構造特徴軸によるキャラクター認識型画像修正ループ

in-house / 2024 / Doc

P34 — From Evaluation to Knowledge: Continuous Learning in Generative Pipelines via Structured Feedback Accumulation

構造化フィードバック蓄積による継続学習

in-house / 2024 / Doc

P35 — A Unified Domain-Specific Language for Heterogeneous Generative Pipeline Composition

異種生成パイプライン統合DSL

in-house / 2024 / Doc

P36 — Multi-Metric Automated Story Evaluation: Beyond Readability to Thematic Coherence via Embedding and Sentiment Analysis

多指標自動小説評価フレームワーク

in-house / 2024 / Doc

P37 — Observability Dashboards for Generative Pipelines: From Score Tracking to 3D Parameter Space Exploration

生成パイプライン観測性ダッシュボード

in-house / 2024 / Doc

P38 — MCP Server Standardization via fastmcp and FastAPI: Transport Unification, SDK Compatibility, and Multi-Protocol Exposure

fastmcp+FastAPI統合によるMCPサーバー標準化

in-house / 2024 / Doc

EvOps Papers — Research Evidence

External Papers

P01 — VeMo: Zero-Shot Text-to-Motion Evaluation using Video Language Models

P02 — EvalCrafter: Benchmarking and Evaluating Large Video Generation Models

P03 — VideoScore: Building Automatic Metrics to Simulate Fine-grained Human Feedback for Video Generation

P04 — VisualPrompter

P05 — 3D Human Animation Quality Assessment: Subjective and Objective Evaluation

P06 — DreamFusion: Text-to-3D using 2D Diffusion

P07 — Point-E: A System for Generating 3D Point Clouds from Textual Descriptions

P08 — Magic3D: High-Resolution Text-to-3D Content Creation

P09 — What matters when building vision-language models? (Idefics2)

P10 — CLIP: Learning Transferable Visual Models From Natural Language Supervision

P11 — Pick-a-Pic: An Open Dataset of User Preferences for Text-to-Image Generation

P12 — ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation

P13 — Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesis

P14 — Training language models to follow instructions with human feedback (InstructGPT / RLHF)

P15 — Direct Preference Optimization: Your Language Model is Secretly a Reward Model

P16 — Active Learning for Large Language Model-based Chatbots

P17 — Model Context Protocol (MCP)

P18 — Agent2Agent Protocol (A2A)

P19 — Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone

P20 — Llama 3.2: Revolutionizing edge AI and vision with open, Customizable Models

P21 — The False Dawn: Reevaluating Performance Gains of Small Language Models

P22 — A Watermark for Large Language Models

P25 — B-CODER: Benchmarking Code Generation with Evaluation Loop

EvOps Original Papers

P23 — VQualA: Video Quality Assessment via VLM

P24 — UniAPO: Unified Automatic Pipeline Orchestration

P26 — MLOps Evaluation Loop Automation

P27 — VLM Evaluation Automation for Visual Content

P28 — Evolutionary Generation Loop with Feedback Accumulation

P29 — Jev-Driven MLOps: High-Speed Judgment Layer for Observable and Controllable Generative Pipelines

P30 — Model-Free Visual Synthesis: Deterministic 3D Rendering via ComfyUI-Workflow Concepts Mapped to Blender-bpy

P31 — Inverse Preference Learning for Evaluation Agents: Inferring Human Value Functions from Scores

P32 — Cost-Aware Rotation Loop Design for Resource-Constrained MLOps: Token Budget Optimization in CPU-Only Generative Pipelines

P33 — Structured Feature Axes for Character-Aware Visual Correction Loops: Multi-Layer Evaluation Combining Detection, Histogram, and VLM

P34 — From Evaluation to Knowledge: Continuous Learning in Generative Pipelines via Structured Feedback Accumulation

P35 — A Unified Domain-Specific Language for Heterogeneous Generative Pipeline Composition

P36 — Multi-Metric Automated Story Evaluation: Beyond Readability to Thematic Coherence via Embedding and Sentiment Analysis

P37 — Observability Dashboards for Generative Pipelines: From Score Tracking to 3D Parameter Space Exploration

P38 — MCP Server Standardization via fastmcp and FastAPI: Transport Unification, SDK Compatibility, and Multi-Protocol Exposure