-
DeepLabV3+ model is plateauing
-
1. Stride 2 conv2d:
`%8 = linalg.conv_2d_nhwc_hwcf_q {dilations = dense : vector, strides = dense : vector} ins(%3, %4, %c0_i32, %c0_i32 : tensor, tensor, i32, i32) outs(%7 : tensor) -> tensor
`
…
-
您好!AMMNet是一篇很优秀的文章,带给我一些思考,想请教一下,为什么segformer的分割性能比DeepLabv3好,但AMMNet训出来DeepLabv3的SSC性能更好呢,难道分割结果不能促进语义场景补全吗?感谢您抽空阅读,您的解答将对我至关重要!
-
The DeeplabV3 i8 model compilation by default might not create the efficient dispatches. Some flags to try
To start with we need `--iree-flow-enable-aggressive-fusion --iree-opt-data-tiling=off`
…
-
hi,how about the performance on pascal voc12 about deeplabv3
xLuge updated
3 years ago
-
Hello, I use resnest50-deeplabv3 model and train from scratch on Cityscapes dataset. During training, the loss value doesn't decrease after reaching 0.1( I use a single titan gpu and batch size of 10…
-
-
Hey! I really loved reading your paper, and was interested in using your approach to assist in some segmentation tasks - do you have a timeframe on when you might release the code for your DeepLabv3+…
-
![default](https://user-images.githubusercontent.com/38428867/50771442-a05fc900-12c5-11e9-8506-861bb46a73c9.png)
![default](https://user-images.githubusercontent.com/38428867/50771469-b3729900-12c5-1…
-
model.py在搭xception时定义了个[activation_fn_in_separable_conv](https://github.com/geng007/models/blob/022c0899a325107821eff6754702ea0f5fb82ed2/fluid/PaddleCV/deeplabv3%2B/models.py#L179) , 我的理解是paper中的Separ…