Comments (14)
also no average pooling is used in the paper if I am not wrong
from caffe-model.
Indeed, the inception resnet v2 implemented by tensorflow is slightly different with the paper. My model is converted from the tensorflow version,but you get the basically same accuracy as the paper.
from caffe-model.
This is a screenshot of part of the tensorflow model
Did you also add a lot of batch norm when the tensorflow model doesn't?
from caffe-model.
I am not very familiar with the operation of tensorflow, but it seems like the batch norm layer is added as follow?
from caffe-model.
Ok. I miss out that part. Thanks a lot. One last question (hope that I am not bothering you) - you don't have the following scaling part in your caffe code right?
I am intending to train the model from scratch.
from caffe-model.
coeff: 1, 0.17 is the scaling parameter.
In my experience, training any inception model from scratch is very difficult, I sincerely recommend you just finetune my model with your dataset.
You can download from BaiduCloud https://pan.baidu.com/s/1jHPJCX4#list/path=%2F
from caffe-model.
Can I know what is the hard part in training inception model from scratch? Do people just keep restarting the training after it dies?
from caffe-model.
Inception model is very hard to optimize, even you follow all the detail as the google's papers say. I don't know why the optimization is so difficult, you can search some relevant repositories to learning some others' experience.
from caffe-model.
Hi I see that your prototxt has a batch size of 64. My Titan X Pascal can only support up to 8 and it is already using more than 7gb. Is that normal?
from caffe-model.
This is normal on caffe, inception-resnet-v2 is memory exhaustion. 64 is just a placeholder when I wrote the script.
from caffe-model.
You say "on caffe". Do you imply that caffe is memory inefficient?
from caffe-model.
Relative to mxnet, it's yes.
from caffe-model.
Hi, just checking. How long did you train your inception resnet model? I am using one pascal titan x and based on my calculation, it is going to take ~6 months for 200 epochs. Is it normal?
from caffe-model.
@yxchng
I just finetuned inception-resnet-v2 for my custom dataset. I forget the specific time consumption, but I remember it's really time-consuming. So taking ~6 months for 200 epoch of imagenet on a single pascal titan x may be normal.
from caffe-model.
Related Issues (20)
- for xception training solver
- inception-v3 get 0% accracy on ILSVRC2012_val.txt? HOT 5
- 你好!问一下目标检测网络R-FCN的问题 HOT 8
- why resnet18-priv outperforms others' implementations so much???
- Can you provide `solver.prototxt` for training pspnet_resnet101?
- Caffemodel files HOT 4
- NASNet caffemodel HOT 1
- multigrid
- ohem
- where is the default location of log file?
- pytorch-cls link is down
- unable to change the size of image size in inceptionv4 model
- issue
- 在哪里可以下载到用于 faster rcnn的DPN预训练模型呢?
- Bug: Incorrect Implementation of Inception_v4 Net in .prototxt file!
- For getting more accuracy in faster rcnn , which parameters i have to tune (tuning parameters)
- Google Drive links broken
- caffemodel without loss layer
- Xception Google Drive link?
- Google drive link broken
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 caffe-model.