xt-models 0.3.1

Models and model utilities for common ML tasks

xt-models

Description

This repo contains common models and utilities for working with ML tasks, developed by Xtract AI.

More to come.

Installation

From PyPi:

pip install xt-models

From source:

git clone https://github.com/XtractTech/xt-models.git
pip install ./xt-models

Usage

Grabbing a segmentation model

from xt_models.models import ModelBuilder, SegmentationModule
from torch import nn

deep_sup_scale = 0.4
fc_dim = 2048
n_class = 2
net_encoder = ModelBuilder.build_encoder(
    arch="resnet50dilated",
    fc_dim=fc_dim,
    weights="/nasty/scratch/common/smart_objects/model/ade20k/encoder_epoch_20.pth"
)
net_decoder = ModelBuilder.build_decoder(
    arch="ppm_deepsup",
    fc_dim=fc_dim,
    num_class=150,
    weights="/nasty/scratch/common/smart_objects/model/ade20k/decoder_epoch_20.pth"
)
in_channels = net_decoder.conv_last[-1].in_channels
net_decoder.conv_last[-1] = nn.Conv2d(in_channels, n_class, kernel_size=(1, 1), stride=(1, 1))
net_decoder.conv_last_deepsup = nn.Conv2d(in_channels, n_class, 1, 1, 0)


model = SegmentationModule(net_encoder, net_decoder, deep_sup_scale)

Grabbing a detection model

from xt_models.models import Model

model_cfg = "./xt_models/models/object_detection/yolov5x.yaml"
model = Model(model_cfg)

Implementing a new model

If you are having to always copy and paste the same model code for different projects, simply add the model code to the models directory, and import it in the models/__init__.py file.

Data Sources

[descriptions and links to data]

Dependencies/Licensing

[list of dependencies and their licenses, including data]

References

[list of references]