Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

What is the minimum GPU memory required to train LResNet100-E-IR on setting MS1M@ArcFace?

See original GitHub issue

@nttstar

I am new to Face Recognition and I have a limited hardware resources. Could you approximately tell us the minimum GPU mem required to train LResNet100-E-IR on setting MS1M@ArcFace? For batch_size of 1 and 128 would be perfect. I currently have 2 GTX 1080s
Could you tell us the approximate performance if I were to train MS1M@Triplet because in case my resources do not suffice for Softmax/ArcFace Traing I will go with MS1M@Triplet

Thank you very much!

Issue Analytics

State:
Created 5 years ago
Comments:5

Top GitHub Comments

1reaction

huyangccommented, Mar 27, 2018

What about the speed of training tripletloss? I’m running at 14.45 samples/secs with 8GPUs @nttstar

0reactions

nttstarcommented, Aug 27, 2018

Tripletloss has extra hard-mining step so speed can not be told simply.

Top Results From Across the Web

Challenging Memory Requirements and Performance ...

Everything we know about memory requirements in machine learning may be wrong. Today, when data scientists process deep learning models ...

face-recognition-resnet100-arcface - OpenVINO™ Toolkit

The original name of the model is LResNet100E-IR,ArcFace@ms1m-refine-v2. Use Case and High-Level Description. Deep face recognition net with ResNet100 backbone ...

GPU memory requirements of a model [closed] - Cross Validated

Suppose your network has 100 uniform layers/feature-maps, each which is 256 by 256 and has 64 channels. This is roughly 420M activation values....

Benchmarking: Which GPU for Deep Learning? - the L7 blog

We already know the best performance/cost GPUs for state-of-the-art deep learning and computer vision are RTX GPUs. So, which RTX GPU should ...

[D] How much VRAM and RAM do I need for NLP transformer ...

I think the bottleneck of training lies in the GPU memory size. I can train bert with 32 sentences every batch on single...