Converting a PyTorch F3Net Model¶
Danger
The code described here has been deprecated! Do not use it to avoid working with a legacy solution. It will be kept for some time to ensure backwards compatibility, but you should not use it in contemporary applications.
This guide describes a deprecated conversion method. The guide on the new and recommended method can be found in the Python tutorials.
F3Net : Fusion, Feedback and Focus for Salient Object Detection
Cloning the F3Net Repository¶
To clone the repository, run the following command:
git clone http://github.com/weijun88/F3Net.git
Downloading and Converting the Model to ONNX¶
To download the pretrained model or train the model yourself, refer to the
instructions in the F3Net model repository. First, convert the model to ONNX format. Create and run the following Python script in the src
directory of the model repository:
import torch
from dataset import Config
from net import F3Net
cfg = Config(mode='test', snapshot=<path_to_checkpoint_dir>)
net = F3Net(cfg)
image = torch.zeros([1, 3, 352, 352])
torch.onnx.export(net, image, 'f3net.onnx', export_params=True, do_constant_folding=True, opset_version=11)
The script generates the ONNX model file f3net.onnx
. The model conversion was tested with the commit-SHA: eecace3adf1e8946b571a4f4397681252f9dc1b8
.
Converting an ONNX F3Net Model to IR¶
mo --input_model <MODEL_DIR>/f3net.onnx