# Customize Object Detection Model Maker

**URL:** <https://discuss.ai.google.dev/t/customize-object-detection-model-maker/30778>\
**Category:** General Discussion\
**Tags:** tflite, coral, help\_request, model\_maker\
**Created:** [November 18, 2021, 9:31pm UTC](https://discuss.ai.google.dev/t/customize-object-detection-model-maker/30778 "2021-11-18T21:31:20Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![heitorrapela](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/heitorrapela/32/9513_2.png) [@heitorrapela](https://discuss.ai.google.dev/u/heitorrapela)\
**Post date:** [November 18, 2021, 9:31pm UTC](https://discuss.ai.google.dev/t/customize-object-detection-model-maker/30778/1 "2021-11-18T21:31:20Z")

</div>

Hello, community!

Is there a way to customize the model from object detection model\_maker? e.g., get the loss and do some early stopping with the validation loss, save and load it. If it is not possible, is there a way to load the model as a tf/keras model to customize the training? Or maybe print the model summary to try to reproduce on tf/keras…

I followed the tutorial with efficiendet, [Object Detection with TensorFlow Lite Model Maker](https://www.tensorflow.org/lite/tutorials/model_maker_object_detection), and it worked fine, besides the fact that the training is being done on CPU.

Best Regards,

---

<div class="post-metadata">

**Author:** ![Mark\_Daoust](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/mark_daoust/32/411_2.png) [@Mark\_Daoust](https://discuss.ai.google.dev/u/Mark_Daoust)\
**Post date:** [November 19, 2021, 6:26pm UTC](https://discuss.ai.google.dev/t/customize-object-detection-model-maker/30778/3 "2021-11-19T18:26:14Z")

</div>

@khanhlvg may have some advice.

But AFAIK model maker is really limited in terms of customizability.

---

<div class="post-metadata">

**Author:** ![khanhlvg](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/khanhlvg/32/7767_2.png) [@khanhlvg](https://discuss.ai.google.dev/u/khanhlvg)\
**Post date:** [November 19, 2021, 11:12pm UTC](https://discuss.ai.google.dev/t/customize-object-detection-model-maker/30778/5 "2021-11-19T23:12:23Z")

</div>

@Yuqi_Li for in-depth advices.

---

<div class="post-metadata">

**Author:** ![Yuqi\_Li](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/yuqi_li/32/7251_2.png) [@Yuqi\_Li](https://discuss.ai.google.dev/u/Yuqi_Li)\
**Post date:** [November 22, 2021, 1:40am UTC](https://discuss.ai.google.dev/t/customize-object-detection-model-maker/30778/7 "2021-11-22T01:40:36Z")

</div>

If you’d like to customize the model, I would suggest you directly use [efficientdet](https://github.com/google/automl/tree/master/efficientdet) repo. Model Maker is based on this repo which limit the customization part.