Where are the model?

[theprodev]

Hi, I see there examples to use squeezenet but I do not see urls to use other models? I want to use coco or yolo? is that possible?

Thanks in advance.

[sleepsonthefloor]

Tensorflow publishes some pretrained object detection models, which you can find here:

https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md

In order to run these with GraphPipe, you need to download the linked .tgz file, uncompress it, and then invoke the model server with a directory mount that allows graphpipe-tf to access the frozen_interface_graph.pb file. For example:

cd /tmp/
wget http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v1_coco_2018_01_28.tar.gz
tar xfvz /ssd_mobilenet_v1_coco_2018_01_28.tar.gz
docker run -it --rm \
        -v "$PWD:/models/"  \
        -p 9000:9000 \
        sleepsonthefloor/graphpipe-tf:cpu \
        --model=/models/ssd_mobilenet_v1_coco_2018_01_28/frozen_inference_graph.pb \
        --listen=0.0.0.0:9000

Now grab a sample image:

wget https://github.com/simo23/tinyYOLOv2/raw/master/dog.jpg

You can call the model with the following example code:

from io import BytesIO
from PIL import Image, ImageOps
import numpy as np
import requests

from graphpipe import remote

data = np.array(Image.open("dog.jpg"))
data = data.reshape([1] + list(data.shape))

pred = remote.execute_multi("http://127.0.0.1:9000", [data], ['image_tensor'],
        ['detection_boxes', 'detection_scores', 'num_detections',
        'detection_classes'])
print("Class predictions: ", pred[-1])

The above code requests four outputs from the model, with the last being detection_classes, which is printed. Note the use of remote.execute_multi, which is a more explicit version of remote.execute. This is the output that I got:

Class predictions:  [[ 18.   3.   2.   8.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.
    1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.
    1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.
    1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.
    1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.
    1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.
    1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.   1.
    1.   1.]]

Class 18 is dog, 2 is bicycle, 3 is car, 8 is truck (https://github.com/nightrome/cocostuff/blob/master/labels.md)

Hope that helps!

[theprodev]

Thanks a lot for the detailed reply. I will check it again soon.