How to train custom object detection model using Google Colab (Free GPU) Part 3

In this video, I am going to teach you how you can use trained model to detect objects in a picture, video or in a webcam.

Website used in this video:

To create a virtual Environment:
conda create -n "Key in name of virtaual environment"

Command used in this video:
pip install TensorFlow==1.15
protoc object_detection/protos/*.proto --python_out=.

[infobea]

I'm just getting the test image without bounding boxes. How can I fix it?

[phoboss]

I have 4 classes with 100 training sets and 50 test sets each is it normal to start of with a verry high loss?
Im at step 21k and my loss value is 22 digits long

[drhollow_on_hogwarts]

thanks, can u please tell how can I build the restful api of the result

[sohailali]

Thanks for the great tutorial. Can you please make a tutorial on how to evaluate trained model in terms of mean average precision mAP?

[gampavenkataprakash]

no such file when running Python setup.py build

[aldrinlewis]

Bro....any idea how to convert this model to tflite???...would be a great help!😅

[ZCVD]

This does not work. If the file Object_detection_image.py MODEL_NAME = 'inference_graph' then nothing happens. If MODEL_NAME = 'new_graph', then an error occurs: Traceback (most recent call last):
File "Object_detection_image.py", line 68, in <module>
tf.import_graph_def (od_graph_def, name = '')
File "C: \ Users \ user \ Anaconda3 \ lib \ site-packages \ tensorflow \ python \ util \ deprecation.py", line 507, in new_func
return func (* args, ** kwargs)
File "C: \ Users \ user \ Anaconda3 \ lib \ site-packages \ tensorflow \ python \ framework \ importer.py", line 431, in import_graph_def
raise ValueError (str (e))
ValueError: NodeDef mentions attr 'incompatible_shape_error' not in Op <name = Equal; signature = x: T, y: T -> z: bool; attr = T: type, allowed = [DT_BFLOAT16, DT_HALF, DT_FLOAT, DT_DOUBLE, DT_UINT8, ..., DT_QINT8, DT_QINT32, DT_STRING, DT_BOOL, DT_COMPLEX128]; is_commutative = true>; NodeDef: {{node GridAnchorGenerator / assert_equal_1 / Equal}}. (Check whether your GraphDef-interpreting binary is up to date with your GraphDef-generating binary.).

[aldrinlewis]

Yeah...while executing install anaconda prompt....it is giving me an error that version cannot be can I do about it??

[sanketrelkar]

I tried to implement using my own images... i wanted two objects to detected... but instead it's only detecting one object and multiple bounding boxes in one image for single object, even when i used the image that contains second object it detected the first object even though it was not in the image... I used only 10 -10 imeges for both the objects to train on google colab, and only gave around half an hour to train them... Could this be reason that it didn't work ? or something else ? Please help me

[colinloh]

Hi there, I cant seem to find the 'training' folder? Is there anything I should look out for?