datur/nncompression

Masters Thesis Project https://davidturner94.github.com/nncompression

Hey I'm Dave

Currently a Backend Developer at Kraken Technologies

nncompression's Issues

identify versions

getElementById can match elements by name instead of ID

---

This issue has been automatically created by todo-actions based on a TODO comment found in docs/build/html/_static/jquery-3.4.1.js:767. It will automatically be closed when the TODO comment is removed from the default branch (master).

model deployment

• deploy a model to cloud
• deply model to edge neural compute stick
• create a module for deployment from onnx to neural compute stick

identify versions

getElementById can match elements by name instead of ID

---

This issue has been automatically created by todo-actions based on a TODO comment found in docs/build/html/_static/jquery-3.4.1.js:781. It will automatically be closed when the TODO comment is removed from the default branch (master).

identify versions

getElementById can match elements by name instead of ID

---

This issue has been automatically created by todo-actions based on a TODO comment found in docs/build/html/_static/jquery-3.4.1.js:767. It will automatically be closed when the TODO comment is removed from the default branch (master).

convert provided onnx model to openvino represnetation

---

This issue has been automatically created by todo-actions based on a TODO comment found in nncompression/convert/to_openvino.py:6. It will automatically be closed when the TODO comment is removed from the default branch (master).

Model evaluation

Create a series of scripts for evaluating and visualizing each model

• model for evaluating openvino model optimizations
• visualize results
• deploy model to cloud and compare with local edge device

add external dependencies

integrate distiller library

• pruning
• quantization

add model compression sub modules

Adding model optimization and compression algorithms

• add pruning algorithms
• add quantization algorithms
• add model optimization

Now that all calls to _data and _removeData have been replaced

with direct calls to dataPriv methods, these can be deprecated.

---

This issue has been automatically created by todo-actions based on a TODO comment found in docs/build/html/_static/jquery-3.4.1.js:4257. It will automatically be closed when the TODO comment is removed from the default branch (master).

Record the results somewhere maybe in a json file or a database or tensorboard

---

This issue has been automatically created by todo-actions based on a TODO comment found in rest_request.py:99. It will automatically be closed when the TODO comment is removed from the default branch (master).

dataset collecting and evaluating

• imagenet
• coco
• cifar10
• cifar100

Experiment with json to binary to deliver the tensor as a list

---

This issue has been automatically created by todo-actions based on a TODO comment found in rest_request.py:98. It will automatically be closed when the TODO comment is removed from the default branch (master).

onnx to OpenVino Conversion

Integrate model conversion from openvino library. This will probably end up calling the mo.py file from the openvino toolkit with the correct flags passed from a file or some other method

• integrate openvino model optimizer module
• find out how to call model optimizer from python function

model architectures

Creating models for evaluation

• vgg
• resnet
• inception
• mobilenet
• efficient net
• yolo
• alexnet
• lenet
• resnext
• squeezenet
• retinanet
• ssd

Add dataloaders to this file

---

This issue has been automatically created by todo-actions based on a TODO comment found in nncompression/models/pytorch/utils.py:211. It will automatically be closed when the TODO comment is removed from the default branch (master).

Now that all calls to _data and _removeData have been replaced

with direct calls to dataPriv methods, these can be deprecated.

---

This issue has been automatically created by todo-actions based on a TODO comment found in docs/build/html/_static/jquery-3.4.1.js:4257. It will automatically be closed when the TODO comment is removed from the default branch (master).

work out accuracy per class

'''

load image into pil format

im = transforms.ToPILImage()(images[0])

convert pil image to bytes

with BytesIO() as output:
im.save(output, 'JPEG')
data = output.getvalue()

url = "http://ml-inference.northeurope.cloudapp.azure.com:5000/inference"

payload = {
'files': (
'1.jpeg',
data,
'image/jpeg'
)
}

r = requests.post(url, files=payload)
r_json = r.json()

outputs = torch.tensor(
r_json['result']['prediction_raw']).unsqueeze(0)

outputs, labels = outputs.to(DEVICE), labels.to(DEVICE)

acc1, acc5 = accuracy(outputs, labels, topk=(1, 5))

top1.update(acc1[0], images.size(0))
top5.update(acc5[0], images.size(0))

measure elapsed time

batch_time.update(time.time() - end)
end = time.time()

if i % print_freq == 0:
progress.display(i)

---

This issue has been automatically created by todo-actions based on a TODO comment found in rest_request.py:101. It will automatically be closed when the TODO comment is removed from the default branch (master).

Average meter for latency inbound and outbound - this should be the time arrived on server minus time sent. and same for response.

---

This issue has been automatically created by todo-actions based on a TODO comment found in rest_request.py:100. It will automatically be closed when the TODO comment is removed from the default branch (master).

identify versions

getElementById can match elements by name instead of ID

---

This issue has been automatically created by todo-actions based on a TODO comment found in docs/build/html/_static/jquery-3.4.1.js:781. It will automatically be closed when the TODO comment is removed from the default branch (master).