laszloC/Thesis_SGM_FPGA

The project is a graduate thesis implementation of a well-known computer vision algorithm (SGM, Hirschmüller 2008). With 18 stars and nearly zero activity for over 7 years, it serves primarily as an academic reference rather than a maintained piece of infrastructure. In the competitive landscape, SGM has been largely superseded by deep-learning-based disparity estimation (e.g., PSMNet, AnyNet) or highly optimized commercial IP cores from vendors like Xilinx (Vitis Vision Library) and specialized silicon from companies like Ambarella or Mobileye. The moat is non-existent as the HLS code is standard for its era and lacks the performance optimizations found in modern vendor-provided libraries. Platform domination risk is low because frontier labs (OpenAI/Google) do not prioritize legacy FPGA vision kernels; however, displacement risk is high due to the shift toward Neural Processing Units (NPUs) and modern transformer-based depth models.