December 2020
tl;dr: Parking slot detection by detecting marking point with a CenterNet-like algorithm.
Overall impression
For my future self: Dataset is super important. Your algorithm is only going to evolve to the level your dataset KPI requires it to.
The algorithm only focuses on detecting the marking point detection and did not mention too much about the post-processing needed to combine the marking points to parking slot. It is more general in that it can detect more than T/L-shaped marking points.
The paper is very poorly written, with tons of sloppy annotation and non-standard terminology.
Key ideas
- A coarse-to-fine marking point detection algorithm. Very much like CenterNet.
- The regression also predicts the “vertex paradigm”. Basically it predicts the pattern of the connectivity among the marking points.
Technical details
- Annotated a dataset (~15k images). This is slightly bigger than PS2.0 dataset with 12k images.
- The paper uses L2 loss to supervise the heatmaps and attributes. This is a bit strange as most studies uses focal loss for heatmap prediction and L1 for attribute prediction.
Notes
- Questions and notes on how to improve/revise the current work