Abstract:
Deep learning with large benchmarks drives object detection development. For real-world applications, collecting fully-annotated data is difficult or expensive, limiting deep neural network power. Humans can use prior knowledge to identify new objects with few elaborately-annotated examples and then generalize this capacity by exploiting objects from wild images.
Our Progressive Object Transfer Detection (POTD) framework is inspired by this learning process. This paper makes three main contributions. First, POTD can use object supervision from different domains for progressive detection.
Human-like learning improves target detection with few annotations. Second, POTD has two delicate transfer stages: Low-Shot Transfer Detection (LSTD) and Weakly-Supervised (WSTD). LSTD distills source detector implicit object knowledge to enhance target detector with few annotations.
Later, it warms WSTD. WSTD uses recurrent object labelling to learn to annotate weakly labeled images. More importantly, we use LSTD’s reliable object supervision to improve target detector robustness in WSTD.
Finally, we run many challenging detection benchmark experiments with different settings. Our POTD outperforms current state-of-the-art methods. https://github.com/Cassie94/LSTD/tree/lstd has the codes and models.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
