Deriving a PyTorch implementation of the YOLOv3 model can be quite challenging. YOLOv3 was NOT originally developed in PyTorch, so it relies on contributions from the community for porting. Many of these ported versions have issues, such as not working as claimed, lacking pretrained weights, or depending on complex dependencies like libGL.so.
To address these challenges, we studied the tt buda demos and adopted the same approach to write YOLOv3 test: Described the model topology (see directory holli_src) ourselves and sourced pretrained weights from external links.
Ticket
None
Problem description
Add model YOLOv3 to trace.
Deriving a PyTorch implementation of the YOLOv3 model can be quite challenging. YOLOv3 was NOT originally developed in PyTorch, so it relies on contributions from the community for porting. Many of these ported versions have issues, such as not working as claimed, lacking pretrained weights, or depending on complex dependencies like
libGL.so
.To address these challenges, we studied the tt buda demos and adopted the same approach to write YOLOv3 test: Described the model topology (see directory
holli_src
) ourselves and sourced pretrained weights from external links.What's changed
Add model test YOLOv3.