It should be ssd-inception-v2, however in the objection detection table it shows ssd-inception-v1 which doesn't work!

It seems trt6.0.1.5 has involved the restriction. In trt5.1.5.0, the ExecutionContext can be destroyed after its corresponding ICudaEngine.

e.g. based on the google doc https://docs.google.com/document/d/1ZI1V_2I3tETAeGnAwYZnrYTdeiOfdtWCL4l4DmRimFc/

p.s. I think that using docker for the course is an overhead which is not justified. As people create a designated VM for the course it would have been better to share a VM image so installation is one click and forget about docker. Docker is great when you need multiple environment or

Describe the bug
The target clean of the root build.sh script should clean all build artifacts of a build made with the standard names and folder paths. Currently it erases almost everything, but fails to delete the cythonized files of the .pyx source files, like python/cuml/cluster/dbscan.pyx. The script should erase those artifacts as well to trigger a "re-cythonization" of all file

From Notebook 5, cell 2

if os.path.basename(os.getcwd()) != 'PConv-Keras':
    os.chdir('..')

from libs.pconv_model import PConvUnet
from libs.util import random_mask, ImageChunker

%load_ext autoreload
%autoreload 2

No such code is found.

I was able to get it to work by using MaskGenerator, I'm assuming you cleaned up the mask generation code then didn't update

It's been a while and I'm still finding it hard to achieve some simple results with OpenCL, either OpenCL C, PyOpenCL or others (tried CUDA). I'd like to try OpenACC, but good luck finding it targeting OpenCL :)

Some general examples that also comment on memory, work items, work groups, queues and atomics if possible (apparently it's not possible to use a mutex or lock up regions if there's sh

In https://github.com/onnx/onnx-tensorrt/blob/master/builtin_op_importers.cpp#L687, there is an assertion for the input types not being kINT32, although it is supported by TensorRT as stated in their documentation:
https://docs.nvidia.com/deeplearning/sdk/tensorrt-developer-guide/index.html#concatenation-layer
"All input tensors must either be non-INT32 type or all must be INT32 type".

Show difference quality as a numerical value between original (full size) image and upscaled image in the waif2x implementation comparison chart.

Probably using SSIM. Would be a informative comparison.

doc under Anakin/benchmark/README_CPU.md is out of data.
Three links of models points to 404 page
Those links in Anakin/benchmark/RNN/prepare.sh are also invalid.

As soon as both futures 0.2 and tokio 0.2 begin to stabilize, create an example which could potentially double as a tutorial and that uses some basic asynchronous features.