This document acts as a sort of to-do list. It contains all the present requirements and features. The plan and order will be followed until the package's completion.
sightseer is centered around 4 major components:
| Component | Description |
|---|---|
| Sightseer | Obtains image data or video footage |
| Proc | Provides image/frame-wise annotation and inter-format conversion tools |
| Zoo | Stores the wrappers over all state-of-the-art models and configs |
| Serve | Provides deployment and model serving protocols and services |
The upcoming version will tentatively contain the following features:
TinyYOLOClient
- Video loading
- Webcam footage
- Screen recordings
- Inter-format data conversion for fine-tuning (XML/CSV/JSON/TFRecords)
- Webcam footage
- Screen grab
- Preloaded video footage
-
xml_to_csv -
json_to_csv -
csv_to_tfrecord -
csv_to_xml -
csv_to_json -
tfrecord_to_csv
Models – Mostly trained on the COCO dataset
- YOLOv3 (
YOLOv3Client) - TinyYOLO (
TinyYOLOClient) - Mask RCNN (
MaskRCNNClient) - Fast RCNN (
FastRCNNClient) - Faster RCNN (
FASTERRCNNClient) - TensorFlow Object Detection (
TFODClient) - Single Shot Detector (
SSDClient) - TensorFlow Object Counting (
TFOCClient) - UNet Image Segmentation (
ImgSegClient)
Add-ons
- Finetuning framework
- Hardware acceleration support (CPU, GPU, TPU)
- Support for TensorFlow 2.0
- Google Cloud Platform support
- Amazon Web Services support
If there are any changes to be made, please submit a PR or file an issue. Once reviewed and finalised, changes can be reflected in the master branch.