Nothing is being detected in Tensorflow Object detection API

Question

I\'m trying to implement Tensorflow object detection API sample. I am following sentdex videos for getting started. The sample code runs perfectly, it also shows the images

Answer 1

The problem is from the model: 'ssd_mobilenet_v1_coco_2017_11_08'

Solution: change to an differrent version 'ssd_mobilenet_v1_coco_11_06_2017' (this model type is the fastest one, change to other model types will make it slower and not the thing that you want)

Just change 1 line of code:

# What model to download.
MODEL_NAME = 'ssd_mobilenet_v1_coco_11_06_2017'

When I use your code, nothing is shown but when I replace it with my previous experiment model 'ssd_mobilenet_v1_coco_11_06_2017' it works fine

Answer 2

I used to have the same problem.

But a new model has been upload it recently 'ssd_mobilenet_v1_coco_2017_11_17'

I tried it and works like charm :)

Answer 3

the function visualize_boxes_and_labels_on_image_array has the following code:

  for i in range(min(max_boxes_to_draw, boxes.shape[0])):
    if scores is None or scores[i] > min_score_thresh:

so, the score must be bigger than min_score_thresh (default 0.5), you can check whether there are some scores bigger than it.

Answer 4

As a workaround change #MODEL_NAME = 'ssd_mobilenet_v1_coco_2017_11_08' to MODEL_NAME = 'faster_rcnn_resnet101_coco_2017_11_08'.

Answer 5

You can use older 'ssd_mobilenet_v1 ... ' and run your program completely with boxes (I run it just now and it is correct). This is a link to this older version. Hope they correct newer version soon!