Javier Berger

@LearnOpenCV Totally agree. I run two detectors in parallel, one generalist, one domain-specific, both can be YOLO or RF-DETR based. IoU match → specific label wins. No match → Gemini Pro labels the unknown for retraining. VLM out of the critical path, where it actually adds value.

Don't use Opus 4.7 for computer vision. It's the wrong tool. I ran a simple pointing task on both Opus 4.7 and GPT 5.4: find the cars in an aerial image. Both took several minutes. That alone should tell you these multimodal models aren't built for this job. Opus 4.7 missed cars entirely, especially cropped ones at the image edges. It also had bad localization (e.g. car 11) and a false positive (car 15) GPT 5.4 caught all of them and it's localization is decent. Neither got close to what a purpose-built detector would produce. If you want to detect cars, use YOLO or RF-DETR. If you need to filter by specific attributes - color, make, orientation - reach for a VLM like Qwen-3VL or Moondream 3. Bottom line: A YOLO model returns an answer in tens of milliseconds. A general-purpose multimodal LLM burns thousands of tokens and makes you wait minutes for a worse result. Use the right tool. Your latency, your cost, and your accuracy will all thank you.

@LearnOpenCV Great post! Completely agree. I actually use this exact workflow: VLMs for rapid bulk pre-labeling to build the dataset, and then I train a fast, low-latency CNN for the actual production deployment. Best of both worlds!

When should you use a Vision Language Model instead of a traditional CNN? CNNs answer structured questions — is there a defect? Where's the pedestrian? VLMs answer open-ended questions using language. Both have their place. If your task is well-defined and repeatable, CNNs still win on speed, cost, and deployment simplicity.

@roboflow Impresionante!!! Gracias por este tremendo regalo de navidad 😊

Announcing Roboflow Rapid: the first prompt-based model creation engine. Data labeling is dead. Go from idea to deployed model in minutes without labeling data. Upload a video, type a text prompt, and get an API. No data labeling teams. No manual annotation. No infra / dependency hell. If you are spending 95% of your time labeling data and only 5% building your app, you are doing it wrong.

#Ahora @robvank en el #IoTDay de la @RoboticMisiones youtube.com/live/3_hqPMrcV…

📚✨ RECONSTRUYAMOS LA UNIVERSIDAD NACIONAL DEL SUR Bahía y la UNS necesitan tu ayuda. Tras la inundación del 7 de marzo cada aporte es vital para reconstruir nuestra universidad. Doná y compartí!! reconstruir.uns.edu.ar

Webinar: Monitoreo inteligente – Cómo la visión artificial revoluciona el agro Inscripciones: shorturl.at/BikYY

Finalizó él RoboChallenge 2024 y el #TeamMisionero se despidió con una destacada participación ‼️ ⏭️ En la Categoría “Line Follower” obtuvieron el puesto N°29 de entre 136 equipos con el robot “Speedy Bee” ⏭️ En Minisumo lograron posicionarse entre los mejores 30 del mundo…

Hoy todos alentando a este Team‼️ Seguilo en vivo por acá: @upb1818?si=uIJFGblhAgXYZTnN" target="_blank" rel="nofollow noopener">youtube.com/@upb1818?si=uI…

Viernes 11: - Panel: “Colaboración Interdisciplinaria: El Rol del Diseño Global” - Simposio: “Visión Artificial: El Poder de Ver Más Allá con Inteligencia” Sábado 12: - Workshop “Construcción de un Detector de Objetos Inteligente para Robótica” Más info: facebook.com/share/p/u57guj…

El 11 y 12 de octubre se realizó el VI Congreso Internacional de Ciencia, Diseño y Tecnología en la @UNAE_Enc y con la @RoboticMisiones desarrollamos varias actividades (+)

A great day for engineering:

The tower has caught the rocket!!

En la @RoboticMisiones impulsamos a los estudiantes para que sean capaces de aplicar estas tecnologías y con las mismas resolver problemas reales, transformar industrias y cambiar el mundo

Estos proyectos no son solo ejercicios académicos, los estudiantes utilizan herramientas que hoy usa la industria en miles de soluciones de IA y esto les permite entender que el potencial de la IA va mucho más allá de aplicaciones de chats para generar texto o imágenes (sigue >)

El Premio Nobel de Física 2024 reconoce a los científicos que facilitaron avances cruciales en el aprendizaje automático y las redes neuronales artificiales. ¿Sabías que estos logros impulsan proyectos innovadores en la @RoboticMisiones🤖🧠 #NobelPrize #IA (sigue >)

BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2024 #NobelPrize in Physics to John J. Hopfield and Geoffrey E. Hinton “for foundational discoveries and inventions that enable machine learning with artificial neural networks.”

🚀YOLO11 Models released by @ultralytics🎉💙 After months of research by Jing Qiu & @GlennJocher, we’re excited to unveil the new Ultralytics YOLO11 models! 📚 Key highlights: ✅ Enhanced architecture ✅ 22% fewer parameters 🔗Learn more: docs.ultralytics.com/models/yolo11/ #AI #ML

@muhammdrizwanmr @ultralytics @GlennJocher

@muhammdrizwanmr @ultralytics

@ultralytics YOLO11 is here!🚀 The wait is finally over! Our latest computer vision model is lighter, faster, and more accurate than ever! 🔥 💥YOLO11 will be available starting Monday, September 30th on Ultralytics HUB and through our pip download package!