search

jkjung-avt / tensorrt_demos

TensorRT MODNet, YOLOv4, YOLOv3, SSD, MTCNN, and GoogLeNet
https://jkjung-avt.github.io/
MIT License
1.75k stars 548 forks source link

GoogLeNet questions #277

Closed krambriw closed 4 years ago

krambriw commented 4 years ago

Hi, I am running your examples since long on my Jetson Nano's, working very well (using YOLOv4, full version)

Now I started to play a bit with the other examples, SSD and GoogLeNet. Basically I got it working following your instructions, no problems at all, but I wonder a few things:

1) Is there another model available for GoogLeNet that is better/able to detect people/pedestrians? The model synset_words.txt has a lot of other stuff but is not really suitable to detect humans (ok baseball player I found) 2) Does GoogLeNet provide data for bounding boxes? 3) If so, do you have a sample python code snippet how to extract and use it from the "out" result (out = net.forward(crop[None])) ?

Best regards, Walter

jkjung-avt commented 4 years ago

GoogLeNet is a image classification model. It only produce 1 "class" output (0~999) for the whole image.

For detecting humans, please refer to my yolov4_crowdhuman. I have included descriptions about Deploying onto Jetson Nano in the README.

krambriw commented 4 years ago

Thank you ver y much for the explanation & links! I am actually very happy with YOLOV, it's really good, so this was just a test of the others. I will with high interest look into the links you provided Best regards, Walter