-
There are 33.6M when I use the 'saved_model' pattern to save the mobilenetv3_large_1.0 models. Is there anything wrong that I can deal with?
-
aimet version: 1.28
SNPE version: 2.14
deploy platform: SM8550 DSP w8 a8 bias32
I have a model, and the backbone of this model is MobileNetV3. As you know, MobileNetV3 primarily consists of poi…
-
Hi Jaehong I recently read your article on Tensorflow: [https://blog.tensorflow.org/2022/06/Adding-Quantization-aware-Training-and-Pruning-to-the-TensorFlow-Model-Garden.html](https://blog.tensorflow.…
-
rvm_mobilenetv3_fp16.onnx 转 TensorRT模型失败,图片中转换报错,请问下有什么办法把AIMatting模型成功转 TensorRT模型吗?
![image](https://user-images.githubusercontent.com/31039387/135009673-f7cbdf1b-5574-490a-9af1-7f7617055c34.png)…
-
Hi guys!
Appreciate your hard work!
I have couple of question that I'd like to clarify:
1. I observe that you have custom implementation of mobile net v3. Did you trained it separately and than use…
-
**System information**.
TensorFlow version: 2.9
Are you willing to contribute it (Yes/No) : Yes
**Describe the feature and the current behavior/state**.
Current `hard_sigmoid` implementation…
-
- 标题:mobileNetv3 onnx转nb(FP16)后,推理输出为nan
- 版本、预测库信息:
1)Paddle Lite 版本:[v2.13-rc](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.13-rc) x2paddle:v1.3.7
2)Host 环境:Ubuntu 18.04
…
-
I have read your codes about split-attention and I found that you use ReLU before split-attention.
https://github.com/zhanghang1989/ResNeSt/blob/76debaa9b9444742599d104609b8ee984b207332/resnest/torch…
-
请问一下,关于mobilenetV3-large 1.00的预训练模型能提供一下吗?
-
raise self.exc_type(msg)
ValueError: Caught ValueError in DataLoader worker process 7.
Original Traceback (most recent call last):
File "/home/yww/anaconda3/envs/pytorch1.3/lib/python3.6/site…