Synthetic Datasets with Blender, Part III

Показать описание

In the last installment of this series, Adam will be addressing accessibility, annotation, automation, and scalability. Annotation is how we go from a bunch of pretty pictures to an actual dataset that can be used to train AI. Automation and scalability are how we make millions of images and accessibility is what opens this process up for everyone to use.

We've just announced a course we're developing (set to release on September 1)

Recommended Links:
Synthetic Datasets with Blender

+ Follow These For Updates +
Subscribe to the YouTube channel too!

+ Connect with Us! +

+ More Ways to Connect +

Рекомендации по теме

Комментарии

As someone who's very interested on using ML for photorealistic VFX, I keep seeing these great things about object detection, face generation, deepfakes and so on but I keep not knowing where to start "learning" it (by "it" meaning the process of actually starting training my own custom models). I thought it'd be easier because I have two decades of coding experience, but even beginner lessons in ML start going really fast into things such as sigmoids, loss functions and other concepts that are either very math-y or very terminological and make me lose track.

Here's a custom use case that comes to mind: if I were to use Google Maps data to recreate a piece of my city in Blender (something I do pretty often for my videos) and wanted an AI to fill that piece of city with things like traffic lights and realistically-moving low-poly cars, where could I get started? Or even going a step further, if for a custom street I wanted to just place low-poly objects in Blender (tagged as "street", "sidewalk" and "storefront" for example) and have the AI first texture them with photographic references, and then fill the scene out with those traffic elements, would this be done with a GAN of some kind? I would love to be able to train models for uses like that since as you know, it's those variations and imperfections that truly sell the realism and they take a ridiculous amount of time to do manually.

Could you point me to the kind of resource you think could be useful for someone in my position? Sorry for the long-winded question.

DodaGarcia

Great video and amazing ideas! Can you do a video on simulating scratches and defects on objects for defect detection using computer vision?

weiyang

can you create one video on how to create all those annotations one by one

RajeshPatil-yclk

Can you make a more elaborated tutorial about how to generate automatically the annotations?

inbard

Can you share source code for image generation in blender?

eduardmart

My computer doesn't run blender 2.8 because It is incompatible with the New OpenGl. Can I make the same with Blender 2.7?

bernardosantannacosta

Synthetic Datasets with Blender, Part III

Synthetic Datasets with Blender, Part I

Synthetic Datasets with Blender, Part II

Synthetic Datasets with Blender, Part III

3D Rendered Datasets in Blender for Beginners Course

Motion Blur & Synthetic Datasets Devlog Series #01

Motion Blur & Synthetic Datasets Devlog Series #02

Will it Blend? Synthetic Datasets with Blender - Decoding Chessboards | Part 3

What is Synthetic Data? No, It's Not 'Fake' Data

Synthetic Data generation from blender

How to make synthetic computer vision datasets in Blender

Synthetic Data Generator for Retail Product Detection.

blender 3D object output bounding boxes to YOLO format for synthetic dataset generation

3D Rendered Datasets in Blender for Beginners, Part 5 [FREE COURSE]

blender human synthetic data

Generating A Synthetic Dataset with Neurolabs

blender human synthetic data 1

Creating data for Deep Learning models using Blender

Synthetic data generation for AI training — Blender Conference 2024

3D Rendered Datasets in Blender for Beginners, Part 1 [FREE COURSE]

Generating Synthetic Datasets with scikit learn

Rendering for Robots (Synthetic AI training data with Blender for computer vision) #b3d

Generating Machine Learning Data in Blender: part 1 - Addon intro

Training a detection network from synthetic data with Blenderproc

Synthetic Datasets: A Missing Link to More Affordable Machine Learning?