Overview¶

The APIs of this package could be divided into the following sub-packages:

The diagram of the MDNC is shown as follows:

flowchart LR
    mdnc:::module
    subgraph mdnc
        optimizers:::blank
        subgraph modules
            conv(conv)
            resnet(resnet)
        end
        models:::blank
        subgraph data
            dg_parse:::modgroup
            sequence(sequence)
            subgraph dg_parse [Dataloaders]
                h5py(h5py)
            end
            preprocs(preprocs)
            webtools(webtools)
        end
        funcs:::blank
        subgraph utils
            tools(tools)
            draw(draw)
        end
        subgraph contribs
            torchsummary(torchsummary)
        end
    end
    classDef module fill:#ffffde, stroke: #aaaa33;
    classDef blank fill:#eeeeee, stroke: #aaaaaa;
    classDef modgroup stroke-dasharray:10,10, width:100;
    classDef ops fill:#FFB11B, stroke:#AF811B;

List of packages¶

optimizers¶

To be built ...

modules¶

• conv: The implementation of the modern convolutional layer and convolutional networks. The networks include U-Net, auto-encoder, encoder and decoder (generator). The codes are inspried by:
• resnet: The implementation of the reisudal blocks and residual networks. The networks include U-Net, auto-encoder, encoder and decoder (generator).

The codes are inspired by the following nice works:

• pytorch/vision/torchvision/models/resnet.py
• milesial/Pytorch-UNet
• nikhilroxtomar/Deep-Residual-Unet

I would like to show my appreciation to them!

models¶

To be built ...

data¶

• sequence: The infrastructures of CPU-based parallel I/O and processing. This module is used by all data loaders.
• h5py: Wrapped HDF5 datasets savers, data converters and data loaders.
• preprocs: Useful pre- and post- processing tools for all data loaders in this package.
• webtools: Web tools for downloading tarball-packed datasets from Github.

funcs¶

To be built ...

utils¶

• tools: Light-weighted recording parsing tools used during training or testing.
• draw: Wrapped matplotlib drawing tools. Most of the utilities are designed as call-back based functions. This module also provide some specialized formatters.

contribs¶

• torchsummary: The revised sksq96/pytorch-summary. This is a Keras style model.summary() in pyTorch, with some bugs gotten fixed. To view my modified version, see sksq96/pytorch-summary!165.

Comments