How to use Tensorflow Object Detection API with multi-channel images

[jrochafe]

Prerequisites

Please answer the following question for yourself before submitting an issue.

• [√] I checked to make sure that this feature has not been requested already.

1. The entire URL of the file you are using

https://github.com/tensorflow/models/tree/r1.13.0/research/object_detection

2. Describe the feature you request

I work with multi-channel images (5 channels). I'm trying to train an object detection model with this images but I haven't found any good support to do it.

I attempt to train a faster RCNN inception resnet v2 network.

3. Additional context

To be more specific, I'm able to modify the input layer ([3, 3, 3, 32] to [3, 3, 5, 32]) in order to adapt one pre-trained model and I'm able to export this model and perform inference with it. For this, I made modifications based on this: https://github.com/minhnhat93/tf_object_detection_multi_channels

The problem comes when I want to train a model using multiple-channel images.

I've tried using the --num_additional_channels option in model_main.py, but I couldn't make it work.

4. Are you willing to contribute it? (Yes or No)

Yes

[DanKlarman]

Following - I am very interested in this, all my attempts so far have failed. Any progress?

[bobleegogogo]

also interested in object detection with multi-channel images. Any updates on the potential feature?