I did not find a single study on U-net-based pedestrian segmentation. Why U-net is inefficient in pedestrian segmentation?
There are works out there that use networks inspired by the U-Net architecture or other architectures. For example, this one Article PedNet: A Spatio-Temporal Deep Convolutional Neural Network ...
is practically the same as U-net with the difference that it takes more inputs and it has more convolutional blocks. The original U-Net proposed by Ronneberger introduced the idea of using long skips to merge high-level and low-level features but the architecture can be improved in many ways; however, nothing stops you from using the original architecture for pedestrian segmentation as shown in this repo: https://github.com/StevenHuang2020/Pedestrian-Segmentation
Thank you so much. I have just started to study segmentation architecture. My goal is to study segmentation-aware pedestrian detection. Therefore, I was looking at popular architecture's merits and demerits.
Dear Sree Nirmillo Biswash Tushar,
UNet is the first ever architecture to study the segmentation of images.It was implemented to segment the tumors among the brain cells . The architecture had both downsampling layers and upsampling layers. The down-sampling layers are usually used to reduce the image to find the global feature of the image, while the up-sampling is used to regenerate the original image.
https://medium.com/@arvindwaskarthik/pedestrian-segmentation-a-study-of-influence-of-parameters-and-datasets-with-unet-716162bac05f
Regards,
Shafagat
Shafagat Mahmudova , Thank you for replying. I read articles about U-net but I do not have detailed knowledge about U-net and therefore, I was more interested in the lackings of U-net that can be improved for pedestrian detection application.
- Badges
- Science topic