-
I don't see an implementation for [depthwise separable convolution](https://arxiv.org/pdf/1610.02357.pdf). Currently it is possible with `Conv2d` by setting `groups=out_channels`. However this is pain…
-
I want to try to use [Mobilenet V3](https://arxiv.org/abs/1905.02244) on some tasks, but there seems to be no official implementation.I know that the architecture of Mobilenet V3 is very similar to MN…
-
The feature_size is the final feature size of mixnet output layer. But, why the value is 1536?Can we change it to other values without affecting the performance of the network, such as 1024, 2048?
-
I wan train mnasnet on GCP using multiple GPU.
So I tried using gcloud ai-platform command.
But it failed.
Could you explain how can i train mnasnet model on multiple GPU machine?
-
I have used GCP VM to run mnasnet using CPU.
I set the parameter "--use_tpu=false" but result message is "RuntimeError: You need to specify a TPU Name if you are running in the Google Cloud environ…
-
Inside the mask_rcnn_main.py,
```
import dataloader
import distributed_executer
import mask_rcnn_model
from configs import mask_rcnn_config
```
Currently if `import sys; print(sys.path)` be…
-
运行这句时
nvidia-docker build -t maskrcnn-benchmark --build-arg CUDA=10.0 --build-arg CUDNN=7 docker/
报错,请问该如何解决?
-
Looks like the structure of the model efficientnet is quite similar to mnasnet?
-
1) In the `mnasnet/utils.py` (master branch), it uses:
```
import sys
sys.path.insert(0, 'tpu/models/official/efficientnet')
import utils as efficientnet_utils
```
Which in my case, it stil…
-
thanks for your great job!
Can you provide pre-training model with mnasnet-a1?