Nevermore

Results 25 comments of Nevermore

In object detection, the image input size is much larger than 224 on ImageNet, you can try FP16 or lower down the input resolution, or choose some platform more suitable...

Not yet. I will test it later.

hi, 看起来你的```out_indices=[1, 2, 8, 9]```应该改成```out_indices=[0, 1, 7, 9]```,不然backbone其实return的4个stage的channel就是[1024, 2048, 4096, 4096],和FPN的输入对不上。

你可以参考这个config的写法https://github.com/huawei-noah/VanillaNet/blob/main/object_detection/configs/vanillanet/mask_rcnn_vanillanet_13_mstrain_480-1024_adamw.py#L26,把init_cfg=dict(type='Pretrained', checkpoint=checkpoint_file)) 放到 backbone的dict中。

没有区别,只是我用的mmdet的版本他们把接口放在了dict里面,我只是调用他们的接口,当然你也可以自己写一个去load ckpt,都是没有区别的。