v0.3.1 android only

Changes

0.3.1

• Using Kivy-Plyer to use native file manager.

v0.2.1 android only

Changes

0.2.1

• Fixing SpeciesNet input data array to have proper detection.

v0.2.0 android only

Changes per version

0.2.0

• Adding SpeciesNet support to identify species from more than 2000 categories. We have converted the PyTorch model to Onnx model.
• Changing the saving directory of raw camera images to output folder, so that user can clean both the raw & output images from Settings.

v0.1.0 android only

Changes

0.1.0

• Everthing till 0.0.5 including the Image Classification feature.
• Setting image preview mode to OFF by default & adding a setting to turn preview ON while browsing the images in file manager (this may slow down or make the app unresponsive).
• Added webp image support.

v0.0.5 android only

Changes per version

0.0.5

• Adding Image Prediction menu which can give top 5 predictions out of 1000 types as per Synset list for the main object on image.
• Changing the model files downlaod style to ask for user prompt.
• Some minor improvements to load the app faster.

Custom Onnx models for this project

Background

We could find many useful models in PyTorch, Tensorflow or many more available formats & we could convert them to Onnx for onnxruntime which can run on low end devices smoothly.

Converted models:

SpeciesNet

This particular model has more than 2000 species around the world including animals, birds etc. More details can be found in their github repo and we do not hold the models license, we just converted it to Onnx

v0.0.4 android only

Changes per version

0.0.4

• Adding a new menu to capture image from camera & detect objects (can have bugs)
• Fix back button action in downlaod manager
• Update relevant links under help section
• Try to gain relevant accesses on Android to perform the tasks, mainly Camera & Image File access

v0.0.3 android only

0.0.3

• Adding download button to download the output image
• Using conditional text size depending on image width

More details are in CHANGELOG

v0.0.2 android only

Vision AI

Vision AI using onnxruntime on python-for-android & kivy. As of now, it has only object detection capability from an image file.