Measured = calibrated with a real training job in this account (3 passes over 980 photos at 512 px). Estimated = from model size, until its first run.
Import a model
From Hugging Face
Any public object-detection model for the transformers library. We check task, licence and size; you approve.
Your own PyTorch model
A .py file defining build_model(num_classes, image_size, pretrained). Download the template.
Waiting for approval
Nothing pending.
Model Registry
SageMaker Model Registry group mf2-wheat-detector. Every model that passed the gate lands here as "pending"; a person approves. The approved one is the champion.
Run inference
Select a model with a running endpoint on the left.
Evaluation images
Training images
Your images
Save any photo from the web and upload it.
expected (drawn by a person)what the model foundClick an image to enlarge.
Results appear here.
New run
The planner only explores the models you pick. Everything here is checked by code before anything trains.