Comments (3)
I submitted a PR to keras-resnet
which is not merged yet, where this repo is depending on (broadinstitute/keras-resnet#27). That would explain the error you are getting. If you want to give it a shot you can try installing this branch of keras-resnet
instead.
As for the completeness, in theory everything is present to train the classification part, the bounding box regression is disabled for now (no use if classification is not working :)). I tried to train a very stripped down version of RetinaNet (computing the class predictions directly on res5, no FPN) on a very simple dataset (a single black image with a white square on it) and got varying results over different runs. Sometimes it would optimize nicely and train a correct classifier, other times it would get stuck at some loss level and it would missclassify a lot. I'm guessing there is some initialization that can be improved.
At the time of writing, the state of this repo is that training is not yet working properly. Contributions are of course always welcome :)
from keras-retinanet.
The RetinaNet paper mentions a new initialization strategy, without which the training does not work, as they say..
There is a PyTorch implementation here https://github.com/kuangliu/pytorch-retinanet, but it is also not complete.
from keras-retinanet.
I tried that initialization but it didn't seem to work for me. It's very well possible that I misinterpreted their paper in some way, but it's difficult to localize where the training is failing.
Btw, the script I'm using to test with is :
https://gist.github.com/hgaiser/c31926937f87ed3a7d5e63aaf130a2b3
from keras-retinanet.
Related Issues (20)
- ERROR:gpu_init.cc(454)] Passthrough is not supported, GL is disabled, ANGLE is HOT 1
- Value Not Accepted: invalid literal HOT 1
- Extracting and saving mAP during training HOT 1
- Training Issue HOT 1
- when training yolov5 on custom dataset on spider show this error? how to resolve this error in pc? HOT 1
- COCO Large category and Anchor boxes generator HOT 1
- Error is displayed when running efficientnet: No module named 'efficientnet'
- Project dependencies may have API risk issues HOT 2
- InceptionV3 error HOT 1
- ModuleNotFoundError: No module named 'keras_retinanet.utils.compute_overlap' HOT 2
- 2023 Model
- Bad mAP results after long train HOT 1
- How to plot the metrics of my custom train
- Error occurred when finalizing GeneratorDataset iterator: FAILED_PRECONDITION: Python interpreter state is not initialized. The process may be terminated. HOT 3
- flag `--patience` (epochs to wait for no observable improvement for early stopping of training)
- How to convert model.h5 to ONNX ? HOT 1
- Error Code 2: Internal Error (ForeignNode does not support data-dependent shape for now.)
- How to get input_1 after training at the input of the model: [?, 512, 512, 3] ?
- train a model with a fixed input size HOT 3
- Not able to run my retinanet on gpu
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 keras-retinanet.