eva-labeling

What is EVA Labeling?

EVA Labeling is a wrapper to enable you to run EVA alongside Label-Studio It lets you connect to Label Studio server to do the following:

• Annotate your dataset more quickly and easily by allowing "bulk labeling" of a set of images.
• Dynamically pre-annotate data based on model inference results.
• Retrain or fine-tune a model based on recently annotated data.

Demo video

bulk-prop-demo.mp4

How it works

1. It first fetches <image, label> pairs from Label Studio server and loads them into the EVA database server.
2. It then runs queries over the loaded data within the EVA database server (e.g., image classification query, object detection query)
3. Finally, it sends the query results (i.e., the updated labels) back to the Label Studio server.

Quickstart

⚠️ The Label Studio server should be started with flag EXPERIMENTAL_FEATURES=1 for the "Bulk Label Propagation" feature to work!

1. Install the dependencies in a virtual environment

   # Install dependencies
   pip install -r requirements.txt

2. Start EVA Labeling Server

   Note: This command will automatically start the EVA server.

   label-studio-ml start ./evaml -eu <eva-server-url> -ep  <eva-server-port> -k <apikey> -ls <label-studio-url>

3. Add the Image Clustering Interface

   <View>
   <Image name="image" value="$image"/>
   <RectangleLabels name="label" toName="image">
     
   </RectangleLabels>
   <TextArea name="cluster" toName="image"
             rows="1" editable="true" />
   </View>

4. Start and Register the EVA ML Backend to Label Studio.

   Following Label Studio ML documentation

   

Adding Custom Models to EVA

1. Register your Feature Extractor model into EVA.

   Refer the EVA documentation for more information. Here is an example of adding a custom model.

2. You can add your custom queries inside ./evaml/cluster_image.py,

   Refer the Label Studio ML Backend documentation for more information.