Yuntao Chen

I can not reproduce this bug. Could you please provide more information about your software and hardware. Does this config run on the COCO dataset?

If you can only run 640 variant, maybe syncbn is not functioning correctly.

Try the pre-built python wheel to ensure you get the right operator.

Here is the wheel for cuda9. https://github.com/TuSimple/simpledet/blob/master/doc/INSTALL.md#setup-locally-with-pre-built-wheel

The stuck may be caused by some image not correctly read in. Is this happening near the end of one epoch.

We pre-flip images in the roidb, so each epoch in simpledet counts for 2 epochs. I can confirm the NASFPN issue and I will look into it soon.

You may download the https://1dv.alarge.space/mask_r50v1_fpn_1x.zip and rename accordingly On Thu, Dec 5, 2019 at 11:03 AM zhangxue123 wrote: > Downloading pretrain_model/r50v1-maskrcnn-coco-0000.params from > https://1dv.alarge.space/r50v1-maskrcnn-coco-0000.params, the html is not > found....

Sorry, this is a compatible issue. A fast workaround should be change `X.conv` to `X.fc` in L470 and L473 https://github.com/TuSimple/simpledet/blob/dafc5536d2ed74094872348801493e42ccaa3c53/symbol/builder.py#L469-L473

Nor yet.

We shall add a pass in the graph optimization for this.