DNN EM segmentation

Training compute

4.8×10¹⁷ FLOP

Parameters

218.9K

Published

Dec 3, 2012

DNN EM segmentation is an AI model developed by IDSIA and SUPSI (Switzerland), first published in December 2012. It works in the vision domain, on tasks such as image segmentation. It counts among the frontier models: the systems trained with the most compute of their moment.

Training it took an estimated 4.8×10¹⁷ FLOP of compute (estimation method: operation counting,hardware). The model has 218,896 parameters. It was trained on roughly 3M datapoints. Training ran on 4 NVIDIA GeForce GTX 580 for about 17 hours. The compute alone is estimated at $4 in 2023 dollars.

Access: Unreleased. Its weights are not openly released. Epoch AI rates the confidence of this record as confident.

Full record

Organization: IDSIA, SUPSI
Country of organization: Switzerland
Domain: Vision
Task: Image segmentation
Training compute: 4.8×10¹⁷ FLOP
Compute estimation method: Operation counting, Hardware
Parameters: 218,896
Dataset size: 3M
Training hardware: NVIDIA GeForce GTX 580
Chips used: 4
Training time: 17 h
Training power draw: 2.1 kW
Training cost (2023 USD): $4
Model accessibility: Unreleased
Open weights: No
Epoch confidence: Confident