deepstream-app and BGR input

Accelerated Computing Intelligent Video Analytics DeepStream SDK
#1

Hi,
I am tying to run inference with the deepstream-app , and I see that when I change the
model-color-format it doesn’t have any influence , and I don’t see this configuration is being parse.
I need to write new deepstream-app if I want to get my input in BGR?

thanks
vadim

#2

Hi,

You should really use this configure to switch RGB/BGR format:

| Property          | Meaning               | Type/Range | Example             |
--------------------------------------------------------------------------------
|                   |                       | Integer    |                     |
| model-colorformat | Color format required | 0 = RGB    | model-colorformat=0 |
|                   | by the model.         | 1 = BGR    |                     |

May I know why you feel the configure doesn’t be passed?
Thanks.

#3

I am running the fasterRCNN sample and I see that when I am editing:
config_infer_primary_fasterRCNN.txt
and change “model-color-format” from 0 to 1 and vise versa the result doesn’t change.
and it can’t be if the color is change from RGB to BGR
thanks

#4

Hi,

Thanks for your update.
We will try this and update more information with you later.

#5

Hi,

Have you deleted the TensorRT engine and recreated it?

rm VGG16_faster_rcnn_final.caffemodel_b1_fp32.engine

If the engine file exist, deepstream will launch TensorRT with the file without recompiling.
Thanks.