Equivalent of min_negatives_per_image

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Please provide the following information when requesting support.

• Hardware (Xavier/Nano)
• Network Type (Yolo_v4 / object detection)
• TLT Version ( tao-toolkit-3-0-21-11)

In pipeline.cfg files when training in tensorflow it is possible to say how many negative examples to use in an image via min_negative_per_image. Is there a similar parameter available in TAO for YOLO specs?

2

Could you share a link for it? Thanks.

3

This from: https://github.com/tensorflow/models/blob/master/research/object_detection/samples/configs/ssd_mobilenet_v1_pets.config

Line 117

       hard_example_miner {
        num_hard_examples: 3000
        iou_threshold: 0.99
        loss_type: CLASSIFICATION
        max_negatives_per_positive: 3
        min_negatives_per_image: 3
      }
4

You shared a link for SSD. So may I confirm that are you checking if there is a similar parameter available in TAO for SSD specs?

5

Thanks for getting back to me. Yes that is for SSD (I don’t think TF1.x has a yolo example), and yes are there similar parameters in TAO either for SSD and/or YOLO?

Many thanks

6

Thanks for the info. TAO does not have similar parameters.

7

Thanks for the clarification

8

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.