Testing Custom Object Detector - TensorFlow Object Detection API Tutorial p.6

Welcome to part 6 of the TensorFlow Object Detection API tutorial series. In this part of the tutorial, we are going to test our model and see if it does what we had hoped. In order to do this, we need to export the inference graph.

[roar]

If you want to do this from outside the models directory, follow my comment on the previous video, then do these:

1) cd into where "training", "data", etc. directories are.

2)Instead of python3 export_inference_graph.py \ (4:25),
run this from that directory: python3 \

2) Instead of running from here (6:31), copy that file into your directory and run jupyter notebook from there.

3) In .ipynb file (6:58) change those two lines under "Object detection imports" to these:

from object_detection.utils import label_map_util

from object_detection.utils import visualization_utils as vis_util

This way utils will be imported from your installation of object_detection module (.egg file), not the object_detection folder in models directory.

4) (*Optional) change TEST_IMAGE_PATH line (8:28) to this so you wouldn't have to rename every test image in your test_images directory:
TEST_IMAGE_PATHS = [ os.path.join(PATH_TO_TEST_IMAGES_DIR, i) for i in

[sifiso]

Sentdex Your tutorials are great, I am a 13 year old programmer with 4 years of experience and with your tutroials, My skills have escalated. Thank you

[xuhaoyi]

Best Object Detection API tutorial so far!
I followed all the steps to train the mac n cheese detector. TF 1.3 also work well if you comment out 3 lines of export_inference_graph.py.
In test, I put some cats and dog images and find that it recognize yellow cats and dogs as macncheese with lower confidence. Van Gogh's starry night won't cause this false positive. I guess put some more categories of images in training dataset will improve the final performance.

[blaublue]

Hi,

this is topic of my bachelor thesis and I have been searching for good basic tutorials since a week. Your tutorial ist the best. I like the speed, not too fast and not too slow. You are so funny too, I was smiling the whole time. Thank you very much! ^^

Best wishes from Germany :)

[sentdex]

So far, I count one macaroni hater in our midst. Stay vigilant. Report all suspicious activity.

[justchill]

Thanks Harrison, I got the ABCD of computer vision from this guy!
Now I can detect cars! Keep up the good work!

[martinzukovec]

Hi sentdex :)

I just finished your tf object detection tutorial. Great job there, i loved it!

I just finished with learning my own model and successfully detected house window on a video. Can't wait to finish my project so that it operates automatically without user input, and i think it will do that after auto tf detection.

Thanks again and keep up good work!

[cristhianandresapolocevall]

Many thanks, man of God! You saved me with a personal project for the university, I really appreciate your work, it opened my eyes to focus
in this subject in my professional career, thank you very much!

[pepajosef]

You man .. are .. I mean thank you so much ! There are no better videos including this topic on youtube than yours. I appreciate your human approach while making it.

[MrSDPace]

Dude! Your TF Object Detection API Tutorials are awesome! Watched a few of them and also read through your corresponding pythonprogramming.net site and was able to train my custom detector using a subset of the coco dataset. Works great on my Linux machine as well as an NVIDIA Jetson TX2!

[janefan]

Thanks for this amazing tutorial series! I hit similar problems as you initially, I'm just glad someone had the patience to figure it out and share with others. If other people are retraining with more than one class, make changes to object-detection.pbtxt, generate_tfrecord and the config file as mentioned, then clear out your old checkpoint files and put a fresh mobilenet copy. Then it will retrain without hitting the tensor shape mismatch error. You have to do this every time you add more classes

[djordjepepic]

Hey man, I just wanna say I love your tutorials, you definitely helped inspire and teach me to become the python programmer I am today! :D

P.S. One thing all of the biggest youtubers do in common is release periodically. I'm not saying commit to a schedule, but I believe you would see much more engagement on your videos from your subscribers if you limited yourself to a single upload per day. Just a thought :) anyways keep being awesome!

[tin]

10/10 - would detect objects again.
I followed along making changes as I went, and got a detection model that works... Only one test image failed to detect, but it was heavily obscured, so I'm honestly not surprised.

[hussamsoufi]

Dude, if you ever come through Berlin let me know, they make the worst macncheese down here! but seriously man, you're a legend, thanks for the Tutorial!

[jpreddy]

Thanks for doing this. Followed your tutorials and created my own model. Otherwise, it would have taken ages for me to figure this shit out.

[GeoRetina]

Hi, thank you so much for this great tutorial. I have been following your videos from the time you started sharing your knowledge on Youtube. Undoubtedly, you're the best.

[justchill]

Thank you very much! :) I don't know who are those 9 people who disliked this informative and successful video! and why?

[lanchens]

love this guy. especially when he laughs.

[miroslavspirkoski]

Wow excellent job Harrison, learned a lot from you. Keep up the wonderful stuff you are doing. I enjoy the series on your channel

[ShahidulAbir]

Eid Mubarak Sentdex and everyone reading this comment.