[TVM Discuss] [Development] [VTA] A Workaround for Deploying Faster R-CNN on Target ext_dev(VTA and ARM CPU)

[Huang Hanting in XJTU via TVM Discuss <no...@discuss.tvm.ai>]

https://drive.google.com/open?id=1io_uQjG9am5mYbFQ-c7h9nH07fYLmVnq

Here is my project including .so files. You can unzip it and run the fasterRCNN_vta.py directly with fsim for vta. I didn't make a git commit because there are some compatibility problems in my code. You can use git status to review my changes. I am going to refactor my code later.





---
[Visit Topic](https://discuss.tvm.ai/t/vta-a-workaround-for-deploying-faster-r-cnn-on-target-ext-dev-vta-and-arm-cpu/6516/8) to respond.

You are receiving this because you enabled mailing list mode.

To unsubscribe from these emails, [click here](https://discuss.tvm.ai/email/unsubscribe/75327e6fbad49477f965f050674d4a49410e3a925a5bcf5f111ffb27f23a3719).

[TVM Discuss] [Development] [VTA] A Workaround for Deploying Faster R-CNN on Target ext_dev(VTA and ARM CPU)

[Hanting Huang via TVM Discuss <no...@discuss.tvm.ai>]

@jinchenglee @acapone13 @Augusto  
I am sorry the shared link cannot be used. I have accelerated the 52-layer convolution on vta in my github dev branch. https://github.com/i24361/incubator-tvm

The consistency problem of vta in zcu 104 platform proved to be an internal logic bug in vta according to https://discuss.tvm.ai/t/rfc-vta-a-hls-c-vta-bug/6743

Due to the characteristics of BRAM, the fallback schedule for vta conv2d causes fault results in real FPGA. There is two way to solve this problem, one is auto-tuning, the other is construct a by-pass for VTA.





---
[Visit Topic](https://discuss.tvm.ai/t/vta-a-workaround-for-deploying-faster-r-cnn-on-target-ext-dev-vta-and-arm-cpu/6516/9) to respond.

You are receiving this because you enabled mailing list mode.

To unsubscribe from these emails, [click here](https://discuss.tvm.ai/email/unsubscribe/d491eb05cd9eb5842adb5913f767ae1d80fca35cb4fde9938245fe275d5e897f).