Comments (4)
Thank you so much for your great work. Could you please provide information about the number and type of GPUs you used for training the models?
a node of 4 A100.
from segvit.
Thank you for your response.
I've been going through your paper and noticed that you used a batch size of 16 for all models. But, when I checked the config files (segvit_vit-l_jax_640x640_160k_ade20k.py), it shows 2 samples per GPU. Given that I'm using 4 A100 GPUs, wouldn't that make the batch size 8?
I'm planning to reference your work and it'd be really helpful if I could get the exact configuration you used. Could you help me out with this?
Thanks in advance!
from segvit.
I believe it refers to a different dataset. In the original answer, we utilized a node with 4 A100 GPUs, with each GPU processing 4 samples in parallel on Pascal dataset due to a smaller crop size. However, if you are training a larger ViT model with a crop size of 640×640, you might require 8 A100 GPUs with 80GB memory, with each GPU handling 2 samples (as done in the current version of SegViT). For the upcoming version, SegViT v2, we will release the code that enables the use of a ViT base model, which will help reduce the cost of reproduction. This release will be available soon.
from segvit.
Thank you so much for your comprehensive response.
from segvit.
Related Issues (20)
- How to train with Vit-Base? HOT 1
- Is there any pre-trained ".pth" file? HOT 1
- Doubt About the Paper: Segmentation Mask HOT 3
- Transfer to BEiT HOT 1
- How to visualize results? HOT 1
- Would you like to provide the pretrained weights? HOT 5
- The mismatch between the model codes and the provided weights HOT 1
- > We trained for around 8 hours for SegViT Large for the Pascal-context dataset with a node of 4 A100. HOT 3
- use checkpointing to save memory HOT 3
- Continual Learning part HOT 3
- random seed HOT 1
- The single-scale result of BEiTv2-Large HOT 1
- Set semseg as logit or probability?
- Asking about the class token. HOT 2
- 模型的权重已无法下载 HOT 2
- Pre-trained weight files missing HOT 1
- Additional loss term for continual learning
- Not able to train the SegVit on the dataset with an input size 512x1024
- SegViTv2 instructions
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from segvit.