Introduction
To promote the development of infrared small target detection technology, we have designed a human-computer interaction system based on a set of model driven algorithms and data driven algorithms, aiming to provide a benchmark platform for researchers to learn, compare, and jointly maintain and update.
The relevant code will be made public later.
In addition, we have also compiled a set of evaluation metrics libraries suitable for algorithms in this field, named BinarySOSMetrics.
The relevant code is published on https://github.com/IRSTD/STD-EvalKit.
The main features of BinarySOSMetrics include:
High Efficiency: Multi-threading.
Device Friendly: All metrics support automatic batch accumulation.
Unified API: All metrics provide the same API, Metric.update(labels, preds) complete the accumulation of batches， Metric.get() get metrics。
Unified Computational: We use the same calculation logic and algorithms for the same type of metrics, ensuring consistency between results.
Supports multiple data formats: Supports multiple input data formats, hwc/chw/bchw/bhwc/image path, more details in ./notebook/tutorial.ipynb