Home
Softono
t

tiger-ai-lab

Professional software vendor delivering innovative solutions on the Softono platform. Specialized in both open-source and proprietary software development.

Visit Website
Total Products
4

Software by tiger-ai-lab

ImagenHub
Open Source

ImagenHub

ImagenHub is a comprehensive Python library designed to standardize the inference and evaluation of conditional image generation models. Recognized at ICLR 2024, the platform defines seven prominent generation tasks and provides curated high-quality evaluation datasets for each. It features a unified inference pipeline supporting over fifty models, enabling fair comparisons across different architectures. The library introduces two specialized human evaluation metrics: Semantic Consistency and Perceptual Quality, accompanied by detailed guidelines to assess generated image quality objectively. ImagenHub includes built-in tools for image visualization, automated metrics, and templates for crowd-sourced human evaluation via Amazon Mechanical Turk. The project is available on PyPI and GitHub, with extensive documentation and a public website showcasing results through ImagenMuseum. It also powers GenAI-Arena, a benchmarking platform for visual generative models. ImagenHub serves researchers and developers by str

AI & Machine Learning ML Frameworks
180 Github Stars
TheoremExplainAgent
Open Source

TheoremExplainAgent

TheoremExplainAgent is an AI system designed to enhance Large Language Model theorem understanding through video-based multimodal explanations. Developed by TIGER-AI-Lab and accepted as an Oral presentation at ACL 2025, the agent generates long-form, animated videos using Manim to visually demonstrate mathematical proofs. This approach helps verify the depth of the model's understanding by uncovering reasoning flaws that often remain hidden in text-only outputs. The project includes the full codebase for generation and evaluation, along with a curated dataset of baseline videos and supporting audio services. It requires a Python environment, specific system dependencies for Manim, and external text-to-speech models. Available resources include the official research paper, a HuggingFace benchmark dataset, pre-generated video data for research baselines, and installation guides for setting up the environment. The system serves researchers aiming to evaluate and improve multimodal reasoning capabilities in AI mo

Media AI Agents ML Frameworks
1.5K Github Stars
VLM2Vec
Open Source

VLM2Vec

VLM2Vec-V2 is a unified open-source framework designed to train and evaluate powerful multimodal embedding models across diverse visual formats, including images, videos, and visual documents. This project introduces MMEB-V2, a comprehensive benchmark comprising 78 tasks tailored to systematically assess embedding performance across these modalities. The framework sets a new state-of-the-art by significantly outperforming strong baselines and specialized models in all evaluation categories. VLM2Vec-V2 features a complete codebase overhaul compared to its predecessor, supporting scalable training and efficient inference for large-scale multimodal applications. The project is developed by TIGER-AI Lab and includes official model checkpoints, datasets, and evaluation tools available on Hugging Face. It welcomes community contributions such as new functionalities, dataset support, bug fixes, and documentation improvements. The software is particularly useful for researchers and developers building retrieval syste

ML Frameworks Data Labeling Vector Databases
652 Github Stars
ClawBench
Open Source

ClawBench

Open-source benchmark for browser AI agents on daily tasks.

AI Agents ML Frameworks
375 Github Stars