Comments (7)
Hi @yashjain-99, yes, we should be adding a full tutorial on transfer learning for image classification soon.
But just for clarification, the json files available under resources contain network descriptions only, not model parameters. For transfer learning, you'd also need the corresponding h5 file from Keras/Tensorflow, which contains the actual model parameters. We will provide that soon as well.
However, I'm curious by what you meant by "it didn't provide any results or return any errors". Were you able to run rmldnn? If so, could you show your config file and what you got as output? Thanks
from rmldnn.
About issue when using same network as given in documentation of MNIST classification but with modification(Attached NN).
rmldnn docker image version: rmldnn latest
Configuration file: config_training.json and config_test.json(attached)
input data file: https://www.kaggle.com/datasets/gpiosenka/100-bird-species
Command run to reproduce the error: sudo docker run -u
rocketml/rmldnn:latest rmldnn --config=config_training.json
Expected behavior: It should have ran and should have given low accuracy as model is not complex enough for this large data.
Screenshots: Attached
Desktop: Ubuntu(22.04) , Docker(20.10.12)
All files available in zip below. Kindly refer that and for dataset use above mentioned kaggle link.
Files.zip
If you require any additional information or files, please let me know.
from rmldnn.
@yashjain-99 could you try the following configuration file? Please note these changes:
resize
indata
-->transforms
to match the input layer in the networkNLL
loss function because this is a classification problem.Dice
loss in your current configuration works for image segmentation problems.
{
"neural_network": {
"outfile": "out.txt",
"num_epochs": 40,
"layers": "./net.json",
"checkpoints": {
"save": "./model/",
"interval": 20
},
"data": {
"input_type": "images",
"target_type": "labels",
"input_path": "./train",
"test_input_path": "./valid",
"batch_size": 8,
"grayscale": false,
"preload": false,
"transforms": [
{ "resize": [224, 224] }
]
},
"optimizer": {
"type": "Adam",
"learning_rate": 1e-4
},
"loss": {
"function": "NLL"
}
}
}
from rmldnn.
Yes, it worked. I will provide the accuracy when the training is finished because it is taking far too long, even though I reduced the number of labels from 400 to 100.
Thanks for help. @ssbotelh @adavanisanti
screenshot attached:
from rmldnn.
An error came up during evaluation: Target -1 out of bound.(Screenshots attached)
from rmldnn.
Hi @yashjain-99, this error is caused by a mistake in the dataset. Notice how the directory
BLACK & YELLOW BROADBILL
has an extra space in the training set, but not in the test or validation sets. As a result, test samples from that category will not have a matching label, thus the invalid -1
target label.
Please rename this directory in the training set by removing the extra space, and retry your run. In future releases, I'll make sure this error happens early and with a clear error message.
from rmldnn.
Yes thanks a lot @ssbotelh , It is working all fine now.
from rmldnn.
Related Issues (4)
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 rmldnn.