microsoft / ell-models

A repository of ELL models

The ELL Models Repository

This is a public repository of pretrained ELL models.

Model Naming

The ELL models repository contains a number of pretrained models suited for different device footprints. These models were generated by parameterizing a few neural network models that are known to work well, and generating a variety of related models based on these starting points. The model files themselves observe the following naming convention:

The filename will start with a short prefix identifying the general model architecture used to train the network. The prefixes are v, d, and a.

Following the prefix will the an underscore (_), followed by an encoding of the input dimension, followed by encodings of the layers present in the model. This encoding uses a single upper-case letter to represent a layerThe layer types and their encodings are:

Layer	Abbrev
Input	I
Convolution	C
Binary convolution	B
Variable-bit conv	V
Max pooling	M
Average pooling	A
Fully-connected	F
Softmax	S

The parameters of the various layers will generally be determined by the architecture, but where necessary, parameters may be added immediately following the layer abbreviation. For instance, a convolutional layer in the 'd' model type uses a 3x3 window for its convolutional layers, except for the last, which uses a 1x1 spatial window. The identifier for a full 'd' model would look like:

d_I224x224x3CMCMCMCMCMCMC1AS

Since many of the models we use alter the structure of the base model they're inspired by, we need to encode the parameters that vary as well. An example of a 'v' model with a single fully-connected layer, and of size 3072, would be:

v_I112x112x3BBMCCMCCCMCCCMCBBMF3072

Code of Conduct

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact [email protected] with any additional questions or comments.

ell-models's Issues

Only getting a 2.86 meg ZIp file. inside model files are 129 bytes.

I can get see a proper featurizer.ell.zip file. unzip or open says damaged file.
I used unblock on the outer zip. other files are text files are fine.
seems to be the case for all models, but only checked Cinnamon and Dulse.

Is this a problem on your end like issue #69 .. or something I am doing.

I went to the upper most level and did a ZIP download.

pretrained models are empty

Hi, I am trying to use the pretrained models, such as it is stored in the following link

https://github.com/Microsoft/ELL-models/raw/master/models/ILSVRC2012/d_I224x224x3CMCMCMCMCMCMC1AS/d_I224x224x3CMCMCMCMCMCMC1AS.ell.zip

(Note: this is copied from tutorial https://microsoft.github.io/ELL/tutorials/Getting-started-with-image-classification-on-the-Raspberry-Pi/ )

But any of the downloaded zip files are corrupted, and the unzip command fails.

Please, could you check?

Thanks in advance.

This repo is missing important files

There are important files that Microsoft projects should all have that are not present in this repository. A pull request has been opened to add the missing file(s). When the pr is merged this issue will be closed automatically.

Microsoft teams can learn more about this effort and share feedback within the open source guidance available internally.

Merge this pull request

Cannot download the raw zip model files.

All the model files in .zip format cannot be downloaded. All I got from the RAW link on the .zip file page is a 133Byte description of the file property.

I even tried to download the whole repository in a zip package. The content of the particular *.zip file is the same 133Byte description.