A lightweight library for importing, manipulating and visualizing images and datasets.
jalabort / nucleus Goto Github PK
View Code? Open in Web Editor NEWA lightweight library for importing, manipulating and visualizing images and datasets.
A lightweight library for importing, manipulating and visualizing images and datasets.
Create the following loss functions:
Commit to master:
Models seem to be unable to be loaded up when using a lambda Layer for dealing with the backbone specific preprocessing.
Replace the custom export function for the public
python package. Functionality is very similar, but public
can also be used for variables.
Need to think about model evaluation. Detection is usually evaluated using mAP
and it would be good to have something very well stablished. Also MODA
has worked well for us in the past. Come up with a plan of action in this regard.
At the moment loading saved datasets is cumbersome, e.g. we need to know their cache and name in advance... This process should be easier and should work in a similar way as it work for model via ModelManager
.
At the moment we have to manually pass our custom objects to the ModelManager
loading functions in order for our models to be loaded correctly. Can we do better and automatically pass them when required automatically?
Separate package in two high level modules:
core
detection
This might end up being two different packages in the future. However, for the time being, it seems development will be easier this way.
Pre-train darknet backbone models on imagenet from scratch.
Adaptive-NMS seems very well suited to our sports-based scenarios. Implement it.
Images are not being displayed the first time we try to visualise something. This is not expected and should be investigated and amended.
nucleus/nucleus/detection/backbones/managers.py
Lines 81 to 147 in 93cbb3d
In some circumstances, I think it would be useful to pass lower-level (higher-resolution) features to a head network through keras.Model
's outputs
kwarg. Selecting the ones you want to expose could probably use something like the current trainable
kwarg pattern.
I don't think we would need to worry about gradient control here, as the head can be configured to pass gradients through to the lower-level weights if needed or not.
Notebooks are difficult to manage with github. However, for lack of a better solution commit the most relevant ones directly to this repo. This will help keeping development overhead to a minimum.
Add the following Keras applications backbones:
MobileNetV2
NASNetMobile
DenseNet121
DenseNet169
DenseNet201
Xception
InceptionV3
ResNetXt50
Have a general pass through fixing, amending and adding docstrings. Give priority to the current detection module.
Can we create a Keras callback that writes images to tensorboard in a similar way as we did in beatrix?
Execute the necessary first steps to set up the project to be used with Polyaxon
.
Think about a more accurate way of representing the relationship between a bounding box score (the score associated to the bounding box being present) and the score associated to the bounding box labels (the scores to the class of object that the bounding box represents).
Create the following detection head layers:
The transform
module is, currently, a bit of a mess. Come up with a good plan for it going forward.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.