A couple of issues in benchmark_score.py

[juliensimon]

Here are a couple of issues in incubator-mxnet/example/image-classification/benchmark_score.py.

Both prevent the script from running.

--> Issue #1
def get_symbol(network, batch_size):
image_shape = (3,299,299) if network == 'inception-v3' else (3,224,224)
num_layers = 0

get_symbol() for vgg16 doesn't accept a zero value for the number of layers.
Quick fix: insert this after "num_layers = 0"
if network=='vgg':
num_layers=16
Or use the same solution as for resnet (split and extract number of layers)

--> Issue #2
devs = [mx.gpu(0)] if len(get_gpus()) > 0 else []

get_gpus() relies on nvidia-smi which is not installed on machines that don't have GPUs (say, c4 instances on AWS).

[goswamig]

Please add labels: "Example", "Need Triage"

[mbaijal]

Please remove the label 'Need Triage'

[szha]

@gautamkmr @sandeep-krishnamurthy please refrain from manually adding "Need Triage" label while triaging. Thanks.

[thomelane]

@juliensimon just tested this out and I don't hit either of these issues.

With regards to Issue 1, there is logic to extract the number of layers from the network string for VGG (as with ResNet).

if 'vgg' in network:
    num_layers = int(network.split('-')[1])

And for Issue 2, I tried get_gpus() on a CPU machine and due to the except OSError exception handling the function just returns an empty list which is expected behaviour of this function. In benchmark_score.py the cpu context is added.

So I think something must have been fixed since you filed this issue. But please do reopen if you're still hitting these issues.

[thomelane]

@indhub @yzhliu would appreciate if you could close this issue, many thanks!