Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

[Support] Wrong input size when converting EfficientDet-lite3

See original GitHub issue

Doing the following to convert EfficientDet-lite3 from https://github.com/google/automl/tree/master/efficientdet:

$ python3 model_inspect.py --runmode=saved_model --model_name=efficientdet-lite3  --ckpt_path=efficientdet-lite3 --saved_model_dir=saved_model/modeldir

$ mo --reverse_input_channels --input_model ../efficientdet/efficientdet-lite3/saved_model/efficientdet-lite3_frozen.pb --transformations_config openvino_env/lib/python3.9/site-packages/openvino/tools/mo/front/tf/automl_efficientdet.json --input image_arrays --tensorboard_logdir . --input_shape [1,512,512,3]

Using the blobconverter webapp to convert the OpenVINO model, tried with both the defaults settings and setting -iml NHWC -il NHWC.

Always the same error when running on OAK-D PRO:

[mxid] [20.1] [168.777] [NeuralNetwork(1)] [warning] Input image (512x512) does not match NN (3x512)

Unclear why this is different from the EfficientDet in the OMZ, which I based this conversion on. What am I doing wrong?

Issue Analytics

State:
Created a year ago
Comments:8 (3 by maintainers)

Top GitHub Comments

2reactions

abroekhofcommented, Sep 2, 2022

Was able to get this working with the following:

mo --reverse_input_channels --input_model ../efficientdet/efficientdet-lite3/saved_model/efficientdet-lite3_frozen.pb --transformations_config openvino_env/lib/python3.9/site-packages/openvino/tools/mo/front/tf/automl_efficientdet.json --input image_arrays --tensorboard_logdir . --input_shape [1,512,512,3] --layout "NHWC->NCHW" --data_type=FP16

I used the default settings for blobconverter in OpenVINO mode.

I guess the NN node in DepthAI expects CHW layout?

0reactions

tersekmatijacommented, Sep 12, 2022

So blobconverter works on either NCHW or NHWC, in fact it is using the compile_tool from OpenVINO in the background, so anything that’s possible to compile with that compiler should be possible to compile with blobconverter. And you’ve managed to compile the blob as you’ve said, but the error was [mxid] [20.1] [168.777] [NeuralNetwork(1)] [warning] Input image (512x512) does not match NN (3x512)

This means that while your model was expecting NHWC, it likely received NCHW images or vice versa. As Erik said, setting colorCam.setInterleaved(bool) should work to my knowledge. Have you tried that?

Top Results From Across the Web

Is it possible to use a custom input shape with efficient det?

Hi,. I want to use TFLite Model Maker to train custom object detection models. When I try to change the hparam image_size.

Input size of converted lite model doesn't match the original ...

Input size of converted lite model doesn't match the original model ... Error (Regular TensorFlow ops) is increasing the app size to 195MB!...

Fail to convert saved EfficientDet model saved_model.pb to ...

As the error suggests, the input tensor should be made up of numbers of 'uint8' data-type. So while constructing the input_data, convert it ......

Retraining EfficientDet for High-Accuracy Object Detection

If the scaled image is smaller than the required input size, the image will be padded with zeros to match the required input...

TFLite Object Detection with TFLite Model Maker

Choose an object detection model architecture. Tensorflow Lite Model Maker currently supports 5 different object detection models (EfficientDet-Lite[0-4]). All ...