Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Cityscapes weights

See original GitHub issue

Hi Vlad,

I am trying to convert the original pspnet101_cityscapes weights from https://github.com/hszhao/PSPNet to keras.

python weight_converter.py pspnet101_cityscapes_713.prototxt pspnet101_cityscapes.caffemodel

In PSPNet-Keras-tensorflow I adjusted:

WEIGHTS = 'pspnet101_cityscapes.npy'
DATA_MEAN = np.array([[[72.78044, 83.21195, 73.45286]]]) # RGB
x = Conv2D(19, (1, 1), strides=(1, 1), name="conv6")(x) # changed the classes from 150 to 19

PSPNet runs through without errors, unfortunately, the output for a test image remains indiscernible:

test out

Do you have an idea what I could be missing?

Issue Analytics

State:
Created 6 years ago
Comments:9 (5 by maintainers)

Top GitHub Comments

1reaction

jmtatschcommented, Oct 19, 2017

@hellochick the original authors used a deeper 101 layer resnet in their pspnet for cityscapes. My mistake was that i tried to convert the weights of a 101 layer network to a 50 layer architecture. Of course the architecture needed a deeper resnet also.

0reactions

hellochickcommented, Oct 19, 2017

@jmtatsch Hey, here is the origin image: test And following image is result produced by your model test_out_seg It looks perfect

Top Results From Across the Web

Detailed Results - Cityscapes Dataset

In this paper we make the observation that the performance of such systems is strongly dependent on the relative weighting between each task's...

class weights of cityscapes #4 - fregu856/deeplabv3 - GitHub

So I add 'torch.cuda.set_device(1)' on the top, but here's a new one 'RuntimeError('cuda runtime error (4) : unspecified launch failure at ' ...

Cityscapes 5000 – Weights & Biases - WandB

Cityscapes fine annotated dataset (no eta reduction). Made by Pier Luigi Dovesi using Weights & Biases.

Why most works on Cityscapes don't use weighted cross ...

Weight Cross-Entroy (WCE) helps to handle an imbalanced dataset, and Cityscapes is quite imbalanced as seen below:.

Envisor® Industrial Rooftop Screens & Enclosures

Envisor® rooftop enclosures & screens from CityScapes, Inc are compatible with many RTU & HVAC systems & meet ... Weight approximately 1.1lbs/sq ft ......