Lightly Blog

How to filter redundant data

Many interesting Deep learning applications rely on the use of complex architectures fueled by large datasets. However, when doing so, a new challenge surfaces: data redundancy.

Efficient Training for Multimodal Vision Models: Techniques and Trade-offs

This article explores the evolution and key design choices in training multimodal vision language models (VLMs). It examines two main architectural approaches: cross-attention (pioneered by Flamingo) and self-attention (used in FROMAGe and BLIP2). We highlight how most modern VLMs build upon pre-trained unimodal backbones rather than training from scratch and discuss various techniques to boost performance, including masked training and resolution adaptation. It also outlines the typical three-stage training process: pre-training, supervised fine-tuning, and alignment, each serving distinct purposes in model development.

Top Computer Vision Tools, Libraries & Frameworks in 2026

A practical guide to the best computer vision tools in 2026, covering deep learning frameworks (PyTorch, TensorFlow, OpenCV), annotation platforms (CVAT, Labelbox, V7), curation tools (LightlyStudio, FiftyOne), pretraining frameworks (LightlyTrain), end-to-end platforms (Roboflow, Encord, Supervisely), and MLOps solutions (W&B, ClearML, MLflow). Includes a quick comparison table and guidance on how to choose the right stack for your ML project.

The Engineer's Guide to Self-Supervised Learning

Learn what self-supervised learning is and how engineers can use it to train AI models with minimal labeled data. This guide explores key techniques, real-world applications, and the benefits of self-supervised learning in computer vision and machine learning.

NVIDIA Blackwell B200 vs H100: Real-World Benchmarks, Costs, and Why We Self-Host

The B200 is up to 57% faster for model training than the H100, up to 10x cheaper to run when self-hosted, and we’ve broken down all the costs, performance metrics, and power consumption data inside.