gpkg.object-detect.models

gpkg.object‑detect.models provides models for each of the architectures support in the gpkg.object-detect base package. Install this package to use these models directly.

If you want to create your own models that extend object detection architectures, refer to gpkg.object-detect for instructions.

Name gpkg.object-detect.models
Description Object detection models (Guild AI)
Version 0.5.1
Source https://github.com/guildai/packages/tree/master/gpkg/object_detect/models
Author Guild AI

faster-rcnn-resnet-101 model

Faster RCNN with ResNet 101.

Operations

detect

Detect images using a trained detector.
Flag Description Default
images Directory containing images to detect. required

evaluate

Evaluate a trained detector.
Flag Description Default
eval‑examples Number of examples to evaluate.

export-and-freeze

Export a detection graph with checkpoint weights.
Flag Description Default
step Checkpoint step to use for the frozen graph.

train

Train detector from scratch.
Flag Description Default
batch‑size Number of examples in each training batch.
clones Number of model clones. This flag has no effect unless `legacy` is `yes`. Set this value to the number of available GPUs for multi-GPU training. . 1
eval‑examples Number of examples to evaluate after training This flag has no effect if `legacy` is `yes` (legacy train does not perform evaluation). .
legacy Use legacy training for object detection Multi GPU support is only available with legacy training. Unlike default training, legacy training does not perform an evaluation after training. . False
quantize Whether or not to quantize model weights. False
quantize‑delay Number of steps to train before quantizing.
train‑steps Number of steps to train.

transfer-learn

Train detector using transfer learning.
Flag Description Default
batch‑size Number of examples in each training batch.
clones Number of model clones. This flag has no effect unless `legacy` is `yes`. Set this value to the number of available GPUs for multi-GPU training. . 1
eval‑examples Number of examples to evaluate after training This flag has no effect if `legacy` is `yes` (legacy train does not perform evaluation). .
legacy Use legacy training for object detection Multi GPU support is only available with legacy training. Unlike default training, legacy training does not perform an evaluation after training. . False
quantize Whether or not to quantize model weights. False
quantize‑delay Number of steps to train before quantizing.
train‑steps Number of steps to train.

Resources

dataset-config

frozen-graph

Frozen inference graph from export-and-freeze.

labels

model-config

  • config/models/faster‑rcnn‑resnet‑101.yml

models-lib

prepared-data

Prepared data for train and validate.

train-config

  • config/train/rcnn‑train‑default.yml

trained-model

Trained model from train or transfer-learn.

transfer-learn-checkpoint

transfer-learn-config

  • config/train/rcnn‑transfer‑learn‑default.yml

faster-rcnn-resnet-50 model

Faster RCNN with ResNet 50.

Operations

detect

Detect images using a trained detector.
Flag Description Default
images Directory containing images to detect. required

evaluate

Evaluate a trained detector.
Flag Description Default
eval‑examples Number of examples to evaluate.

export-and-freeze

Export a detection graph with checkpoint weights.
Flag Description Default
step Checkpoint step to use for the frozen graph.

train

Train detector from scratch.
Flag Description Default
batch‑size Number of examples in each training batch.
clones Number of model clones. This flag has no effect unless `legacy` is `yes`. Set this value to the number of available GPUs for multi-GPU training. . 1
eval‑examples Number of examples to evaluate after training This flag has no effect if `legacy` is `yes` (legacy train does not perform evaluation). .
legacy Use legacy training for object detection Multi GPU support is only available with legacy training. Unlike default training, legacy training does not perform an evaluation after training. . False
quantize Whether or not to quantize model weights. False
quantize‑delay Number of steps to train before quantizing.
train‑steps Number of steps to train.

transfer-learn

Train detector using transfer learning.
Flag Description Default
batch‑size Number of examples in each training batch.
clones Number of model clones. This flag has no effect unless `legacy` is `yes`. Set this value to the number of available GPUs for multi-GPU training. . 1
eval‑examples Number of examples to evaluate after training This flag has no effect if `legacy` is `yes` (legacy train does not perform evaluation). .
legacy Use legacy training for object detection Multi GPU support is only available with legacy training. Unlike default training, legacy training does not perform an evaluation after training. . False
quantize Whether or not to quantize model weights. False
quantize‑delay Number of steps to train before quantizing.
train‑steps Number of steps to train.

Resources

dataset-config

frozen-graph

Frozen inference graph from export-and-freeze.

labels

model-config

  • config/models/faster‑rcnn‑resnet‑50.yml

models-lib

prepared-data

Prepared data for train and validate.

train-config

  • config/train/rcnn‑train‑default.yml

trained-model

Trained model from train or transfer-learn.

transfer-learn-checkpoint

transfer-learn-config

  • config/train/rcnn‑transfer‑learn‑default.yml

pet-images model

Annotated images from Oxford-IIIT pet dataset.

Operations

prepare

Prepares pet images for training.

This operation does not have any flags.

Resources

ssd-mobilenet-v2 model

SSD with MobileNet v2.

Operations

detect

Detect images using a trained detector.
Flag Description Default
images Directory containing images to detect. required

evaluate

Evaluate a trained detector.
Flag Description Default
eval‑examples Number of examples to evaluate.

export-and-freeze

Export a detection graph with checkpoint weights.
Flag Description Default
step Checkpoint step to use for the frozen graph.
tflite Whether or not to export graph with support for TensorFlow Lite. False

train

Train detector from scratch.
Flag Description Default
batch‑size Number of examples in each training batch.
clones Number of model clones. This flag has no effect unless `legacy` is `yes`. Set this value to the number of available GPUs for multi-GPU training. . 1
eval‑examples Number of examples to evaluate after training This flag has no effect if `legacy` is `yes` (legacy train does not perform evaluation). .
legacy Use legacy training for object detection Multi GPU support is only available with legacy training. Unlike default training, legacy training does not perform an evaluation after training. . False
quantize Whether or not to quantize model weights. False
quantize‑delay Number of steps to train before quantizing.
train‑steps Number of steps to train.

transfer-learn

Train detector using transfer learning.
Flag Description Default
batch‑size Number of examples in each training batch.
clones Number of model clones. This flag has no effect unless `legacy` is `yes`. Set this value to the number of available GPUs for multi-GPU training. . 1
eval‑examples Number of examples to evaluate after training This flag has no effect if `legacy` is `yes` (legacy train does not perform evaluation). .
legacy Use legacy training for object detection Multi GPU support is only available with legacy training. Unlike default training, legacy training does not perform an evaluation after training. . False
quantize Whether or not to quantize model weights. False
quantize‑delay Number of steps to train before quantizing.
train‑steps Number of steps to train.

Resources

dataset-config

frozen-graph

Frozen inference graph from export-and-freeze.

labels

model-config

  • config/models/ssd‑mobilenet‑v2.yml

models-lib

prepared-data

Prepared data for train and validate.

train-config

  • config/train/ssd‑train‑default.yml

trained-model

Trained model from train or transfer-learn.

transfer-learn-checkpoint

transfer-learn-config

  • config/train/ssd‑transfer‑learn‑default.yml

voc-2008-images model

Visual Object Classes Challenge 2008 images.

Operations

prepare

Prepare images annotated using Pascal VOC format.
Flag Description Default
random‑seed Seed used for train/validation split.
val‑split Percentage of images reserved for validation. 30

Resources

dataset-config

  • dataset\.yml from prepare operation

labels

  • labels\.pbtxt from prepare operation

models-lib

prepared-data

Prepared data for train and validate.
  • .*record.* from prepare operation

voc-annotated-examples

voc-annotated-images model

Images annotated using Pascal VOC format.

Operations

prepare

Prepare images annotated using Pascal VOC format.
Flag Description Default
annotations Directory containing image annotations. required
images Directory containing images to prepare. required
random‑seed Seed used for train/validation split.
val‑split Percentage of images reserved for validation. 30

Resources

dataset-config

  • dataset\.yml from prepare operation

labels

  • labels\.pbtxt from prepare operation

models-lib

prepared-data

Prepared data for train and validate.
  • .*record.* from prepare operation