Getting started with CREStereo

I would like to trying running a stereo ML model called CREStereo on the Hailo8 (GitHub - megvii-research/CREStereo: Official MegEngine implementation of CREStereo(CVPR 2022 Oral).). I used the DCF_1_Parsing_Tutorial as my starting point. However, at the first stage of parsing the ONNX into HAR, the ClientRunner warns that this is a large model, and then crashes after 30 minutes with “IndexError: list index (0) out of range” in hailo_sdk_client/model_translator/onnx_translator/onnx_graph.py: line 772. Doing some more digging I found that this ONNX file has 27791 vertices in the computational graph, well above the large model threshold of 1500. The ONNX file itself is not very large at 25MB, considerably smaller than the 46MB for resnet_v1_18.onnx that is in the tutorial. We are able to run this model on an Nvidia Jetson AGX at 10 fps so I thought it might work on Hailo8. Please advise me if there are some further steps I can take.

Hi @alsulli,

Welcome to the Hailo Community!
We are looking into this and will get back to you.