Lightly

Improve machine learning models by curating vision data. Join our Discord

[Lightly @LightlyAI]

Big news 🎉 We’ve just released LightlyStudio, our new open-source platform for multimodal https://t.co/8SuAzY0Iqo can now curate, label and manage data all in one place.

🔗 Read more: https://t.co/HhaJSzv2UU
💻 GitHub: https://t.co/DYTPaj5HJV

[Lightly @LightlyAI]

We're at #ICRA2026 in Vienna this week 🇦🇹🎉 If you're building robots and thinking seriously about perception - come find our co-founder and CTO Igor Susmelj. He's here all week and happy to talk computer vision and how Lightly fits into your stack.
👉 https://t.co/NJ01UZg0XX

[Lightly @LightlyAI]

Excited to welcome Gabriel Fruet to the team!

Gabriel joins us as a Machine Learning Engineer, working on LightlyTrain and LightlySSL. Really glad to have you with us 👋

[Lightly @LightlyAI]

🤖 We're going to ICRA 2026 in Vienna (June 1-5)!

Our Co-Founder & CTO Igor Susmelj will be there. Let's grab a coffee and chat about robotics, perception, and data. ☕

Will you be there? Drop a comment or reach out! 👇

[Lightly @LightlyAI]

Your data curation logic is your competitive advantage.

pip install lightly-studio

Start locally. No account, no data leaving your infra. Full control over the curation layer.
→ 🔗 https://t.co/pqna7IGQnt

[Lightly @LightlyAI]

We're happy to welcome Nauryzbay to the Lightly team joining us as an ML Engineer on LightlyTrain - working to make state-of-the-art computer vision training more accessible.

Great to have you with us 👋

[Lightly @LightlyAI]

Real-world datasets are messy in more than one way - bad metadata AND visual issues.

Most tools handle one or the other.

LightlyStudio combines metadata filtering + semantic search in a single query. 💡

🔗 https://t.co/ZPRNQ8jD1d

[Lightly @LightlyAI]

Most teams treat data curation as a one-time cleanup job.

It's not. The best ML teams treat it as a loop: Curate → Label → Train → Evaluate → Mine → Refine.

🔗 Check it out on GitHub: https://t.co/YMasWoEPV6

[Lightly @LightlyAI]

Big welcome to Nicholas Koelln, who's just joined us as a Machine Learning Engineer!

Really glad to have you on the team, Nicholas 👋

[Lightly @LightlyAI]

LightlyTrain 0.15.0 is out 🎉
✅ Distillationv3 - one method for classification AND detection/segmentation
✅ Meta EUPE backbone support (ViT + ConvNeXt)
✅ Custom teacher models
✅ Mosaic, MixUp & CopyBlend augmentations
👉 https://t.co/3VFjkhZ7ZV

[Lightly @LightlyAI]

Can't describe what you're looking for - just drop an image.

LightlyStudio's multimodal search returns similar results instantly, across your entire dataset. No metadata. No labels needed.

🔗 Try it on your own data → https://t.co/6ApjF2rhsb

[Lightly @LightlyAI]

We're heading to CVPR 2026 in Denver ☀️⛰️

Come find us at booth 514 - we'll have limited edition swag and our DINO ducks waiting for you.

If you're working on computer vision data pipelines, let's talk. Book a demo ahead of time: https://t.co/8Xgq9qjwVQ

[Lightly @LightlyAI]

🧠 “We found the code at @LightlyAI to be super well written and it was easy for us to make modifications.”

Surai Pai from Harvard Medical School shared why his team used LightlySSL to adapt DINOv2 for 3D CT segmentation.

📄 Read more: https://t.co/4eR69GAWGK

[Lightly @LightlyAI]

🎥 Eurocontrol X Lightly: Ana-Maria Pelin explains how her team detects contrails from satellite imagery, and why they turned to LightlyTrain to improve segmentation performance with YOLO on ash-RGB data.

Watch the clip 👇

[Lightly @LightlyAI]

🇮🇹 This week, the Lightly team is in Tuscany - workshops, strategy sessions, and team time just outside Florence.

Afternoons = Italian food, winery visits, and dreamy views 🍷🌄

Heading home tomorrow recharged, aligned, and ready to make 2026 our most ambitious year yet. 🚀

[Lightly @LightlyAI]

🚀 LightlyTrain 0.14.0 is out: smaller and faster vision models for real-time deployment.

Read more: https://t.co/eUIzax6tbJ

[Lightly @LightlyAI]

What does it take to train a 3D foundation model for CT segmentation?

Surai Pai from Harvard Medical School shares how his team used LightlySSL to adapt DINOv2 for volumetric inputs, unify training configs, and streamline workflows across researchers.

🔗 https://t.co/lA6KvSLUiG

[Lightly @LightlyAI]

How do you train vision models on satellite imagery where RGB pretraining fails?

EUROCONTROL used LightlyTrain for contrail detection, boosting YOLO performance with distillation + unlabeled data. Results are preliminary but promising.

Read more: https://t.co/KMa6nl67vD

[Lightly @LightlyAI]

We launched plugins in LightlyStudio last month and the feedback has been great. 🤩

They give ML teams a clean way to embed custom logic and automation directly into their dataset workflows.

Read more → https://t.co/vTlUYFaYUd

[Lightly @LightlyAI]

A great evening celebrating our year with the Lightly team at our Christmas dinner.🎄🌟

We’re grateful for every person on this team and for the dedication and passion they bring.

We're looking ahead to what we will achieve together in 2026.🎉 Big things are coming!

[Lightly @LightlyAI]

📹 LightlyStudio now supports richer multimodal data workflows with full video and caption integration, including Python video loading/exploration, instant hover-over previews, and UI tools for viewing, editing, and QAing multiple captions.

Read more: https://t.co/JgNxfPBUhV