How to Build an Object Detection Classifier with TensorFlow 2.0 on Windows/Linux

[@TheAIGuy]

Hi everyone! I have updated the code to allow for webcam detections as well! Refer to the readme to see how! github.com/theAIGuysCode/yolo-v3

[@matkoschlebikom3630]

hello i need work witch coordinates object , how i extract coordinates from yolo_v3.py

[@Lasheen419]

Good info, could you briefly run through the code in future videos. Maybe just a 2-3 overview of what each file you are using does. This will greatly help everyones learning. Thanks

[@antonermakow9232]

The easiest guide I've seen. Thanks a lot for working. Subscribed, like.

[@TheAIGuy]

Hi everyone! Let me know if you run into any issues along the way in the comments below. Keep posted for updates on the code!

[@rohitborra2507]

everything is fine(detection with images and webcam) but I not able to do it with video please help (I have subscribed your channel ).

[@tomberry1106]

Please help me. When I launch it the second time: W tensorflow / core / framework / op_kernel.cc: 1622] OP_REQUIRES failed at save_restore_v2_ops.cc:184: Not found: Key yolo_v3_model_1 / batch_normalization_100 / beta not found in checkpoint Exception in thread Thread-7:

[@erolgyte]

cannot detect

[@tejassawant2957]

Is it necessary to have a gpu to run this prog?

[@TheAIGuy]

@@tejassawant2957 not necesssary but will help speed up the program by a decent amount!

[@GagarineYuri]

Hi, thanks for the howto. Just one thing, you need to set the numpy version to 1.19.5 in the requirements in order to make it work (the load_weights.py raise a NotImplementedError)

[@DroneShorts360]

You are Lifesaver !!! Thanks

[@appxpress]

Thanks it helped me alot

[@JHauris]

These are great tutorials - very well done. Thanks for doing them. I would be very interested in a video on how to train a custom data set from within tensorflow (rather than via darknet). Also videos on PPO (Proximal Policy Optimization) in reinforcement learning would be very cool. Thanks, Jon

[@TheAIGuy]

Thanks! I will keep these in mind going forward.

[@CJ-wk1xj]

@@TheAIGuy please let us know when you make a video on it :)

[@satishpoojith4709]

Excellent video it is working Perferctly it is a great job eagerly waiting for you next video "object detection with help of cloud GPU"

[@TheAIGuy]

Thanks so much! I will be making a YOLOv3 in the cloud with GPU soon!

[@TheAIGuy]

Yolov3 in the Cloud with Free GPU is now uploaded! Check it out !

[@MrCreatorCC]

Very informative thanks for sharing

[@evelynjunco4574]

Thank you for the tutorial! you are using MINGW64, is that the same as Command Prompt? I am not familiar with it and I'm having trouble downloading it. Any suggestions?

[@TheAIGuy]

Evelyn Junco command prompt works just the same. There is no need for mingw!

[@tomberry1106]

Please help me. When I launch it the second time: W tensorflow / core / framework / op_kernel.cc: 1622] OP_REQUIRES failed at save_restore_v2_ops.cc:184: Not found: Key yolo_v3_model_1 / batch_normalization_100 / beta not found in checkpoint Exception in thread Thread-7:

[@yadnik2016]

ERROR: Could not find a version that satisfies the requirement tensorflow==2.0 i am getting this error when i am installing requiremnts file. what should i do

[@timverlinde8954]

Did you recieve an answer or found a solution?

[@dG7DraGoSs]

@@timverlinde8954 @YADNIK did you find solution?

[@rohitpandey9920]

PLEASE HELP!! when i ran 'pip install -r requirements.txt', i got the following error :- 'ERROR: Could not find a version that satisfies the requirement tensorflow==2.0 (from versions: 2.8.0rc1, 2.8.0, 2.8.1, 2.8.2, 2.8.3, 2.8.4, 2.9.0rc0, 2.9.0rc1, 2.9.0rc2, 2.9.0, 2.9.1, 2.9.2, 2.9.3, 2.10.0rc0, 2.10.0rc1, 2.10.0rc2, 2.10.0rc3, 2.10.0, 2.10.1, 2.11.0rc0, 2.11.0rc1, 2.11.0rc2, 2.11.0) ERROR: No matching distribution found for tensorflow==2.0'

[@arfadk6291]

Hello AI Guy, Your whole tutorial on object detection is awesome. Well organised and clearly structured so that any beginner can follow. I'm doing my final year project on detecting vehicles and number plates from low end cameras (indoor security cameras). What I plan to do is run the custom object detector for vehicles on the live video first, then crop the image of the vehicle and then detect the number plate from that(I want the type of the vehicle too that is why this approach). But the problem is how could I select which frame of the vehicle to crop so that i can get a clear number plate for the further detection and recognition of the number plate. Could you suggest any method to extract the best frame during object detection? I'm stuck with this problem,so your suggestion would be a great help Thanks

[@MrDEROS7]

Hi, I'm also working on something similar for my final year project, have you figured out how to do it? Thanks

[@arnavmehta3669]

Atlast found the right one after hours!

[@syazwanasyraf6567]

Heyyy AI Guy, your tutorial is super great. The only tutorial that I successfully run since I'am still new in computer vision. I have one question, how can I produce the result of Frame per second (FPS) for the video and webcam? What should I add? I really hope you can help me. Thank you

[@TheAIGuy]

Syazwan Asyraf glad you enjoy it! Can you give me a little more detail on what your are wanting to have done? You just want to output the FPS you are getting on the video?

[@syazwanasyraf6567]

@@TheAIGuy yes that is what i mean. I just want to get the result of the FPS😅. Can you help me?

[@ridael-mehdawe4681]

thanks for the tutorial, but it can NOT detect traffic lights, unless the traffic light are really big in the image... should I do training for that class? any info will be appreciated. peace

[@TheAIGuy]

Rida El-Mehdawe you can always make a custom detector for sure. You will just need lots of training images of traffic lights.

[@ridael-mehdawe4681]

@@TheAIGuy thanks for your prompt response... I got confused, cz there is a pre-trained model called "faster_rcnn_resnet101_coco_11_06_2017 " this one even smaller in size than yolov3, but it DETECTS traffic light. Does that mean yolo is NOT the best? if so, may I know the best pre trained weights. peace

[@thealgorithm7633]

I want to detect electronic gadgets like camera, mobiles and tablets. Is there any similar weights available like coco weights? or do I need to create a custom weights?

[@TheAIGuy]

You will have to do some research, COCO has phones, computer etc but maybe not tablets. Custom weights can be easily created using my YOLOv3 in the cloud video.

[@mogivara]

Can you help? I got such an error when I was processing python detect.py video 0.5 0.5 data / video / shinjuku.mp4 video with this script, I don’t understand what’s wrong here, everything is working fine with the pictures,TF 2.0.0,all that was in the description, uploaded Traceback (most recent call last): File "detect.py", line 146, in main(sys.argv[1], float(sys.argv[2]), float(sys.argv[3]), sys.argv[4:]) File "detect.py", line 62, in main cv2.namedWindow(win_name) cv2.error: OpenCV(4.2.0) C:\projects\opencv-python\opencv\modules\highgui\src\window.cpp:634: error: (-2:Unspecified error) The function is not implemented. Rebuild the library with Windows, GTK+ 2.x or Cocoa support. If you are on Ubuntu or Debian, install libgtk2.0-dev and pkg-config, then re-run cmake or configure script in function 'cvNamedWindow'

[@TheAIGuy]

Seems like an opencv issue which is strange, maybe try a newer version of opencv.

[@mogivara]

@@TheAIGuy opencv v4.2.0 )

[@mogivara]

@@TheAIGuy maybe I need a version that was in development?

[@seb4stian5]

Excellent video! Very well explained

[@DanielWeikert]

loading the weights worked but when trying to detect I run in a dll load failed error C:\Users\weikdan\Desktop\yolo\yolo-v3>python detect.py images 0.5 0.5 data/images/dog.jpg data/images/office.jpg 2020-01-23 09:09:40.710587: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_100.dll Traceback (most recent call last): File "detect.py", line 20, in from utils import load_images, load_class_names, draw_boxes, draw_frame File "C:\Users\weikdan\Desktop\yolo\yolo-v3\utils.py", line 4, in from PIL import Image, ImageDraw, ImageFont File "C:\Users\weikdan\AppData\Local\Continuum\anaconda3\lib\site-packages\PIL\Image.py", line 94, in from . import _imaging as core ImportError: DLL load failed: Das angegebene Modul wurde nicht gefunden. All modules are installed and can be imported in e.g. jupyter nb any ideas?

[@TheAIGuy]

Daniel Weikert it seems to be an issue with your imported Pillow package. I would try uinstalling pillow and reinstalling it.

[@erolgyte]

@@TheAIGuy i got the same error

[@michaelroditis1952]

Maybe it's a stupid question but is the train.txt file the custom weights file?

[@TheAIGuy]

Michael Roditis the custom weights file is a .weights file type. The train.txt is just a list of all the paths to training images.

[@Mirror_Holder]

Hello the AI guy, thanks for the video. I'll be doing my final year university project next semester. It will be a device to help visually impaired and what I want to do is to use YOLO with some proximity sensors and output audio for the user. What single board computer do suggest I use and what do you think is the best approach to output the detected thing as an audio

[@TheAIGuy]

You can perform YOLO on a raspberry pi and also have a microphone and speaker attached to the raspberry pi through bluetooth. That would be my recommended approach!

[@kushagraanand7321]

Hello! Thanks for this wonderful video! I am trying to make a human fall detection model and I am thinking of just detecting a person in a video and no other object can you please tell how?

[@mellarksgirl]

i'm really new to all of this so this is gonna sound dumb but what does custom weights mean? please help. is it training your own datasets?

[@zhengduozhu3865]

Hi AI guy~ Your video is clear and useful, I just want to know if you can pre-training your data and increase the detection speed?

[@sectember]

Hello :) could you recommend me a tutorial to integrate this project in a web app with Django or something like this? By the way, im thankfull for this one :)! Cheers from Brazil!

[@michaelroditis1952]

I installed the packages but still when I run it from the terminal I get ModuleNotFoundError. When I run it from pycharm everything is alright but I don't know how to add parameters at detect.py file without changing the code

[@TheAIGuy]

Try using the code from my new video "Realtime Yolov3.." it is optimized for tf2 and is easier to customize!

[@eyronethanoel2266]

Dudeee i subscribeeeeed. Cant wait your next video

[@하진석-t6f]

Thank you for video. It is very helpful! I have a question about the model config file. In that file, there are some lines like eval_config, and eval_input_reader. Those eval means validation during the train? or test the model?

[@hajeratitallah8936]

Very interesting video, what changes should i make to the code in order to train my own models. thank you

[@TheAIGuy]

Great question. Since the weights are converted from yolo weights, you can use my "Yolov3 in the Cloud" or "Train Yolov3 Custom Object Detector" to train your own yolo weights and then convert them in to the TensorFlow model. I am currently working on code to train in the TensorFlow model.

[@FarzanMNoori]

Hi, I have trained my custom model on Darknet original repository having YOLO-V3. But your Tensorflow2.0 version is super faster than the original. How can I use a *custom trained model* on your code?

[@teetanrobotics5363]

Amazing tutorial bro.

[@xaeroekis]

Hey do you plan on doing a tutorial on how to train a SSD object detection with the new tensorflow 2 object detection API? Ive seen like no good tutorials on the new API.

[@joooooojo2235]

Hi, I would like to ask what is the performance difference between using this installation and the darknet way of installation? This seems much less prerequisite installation than the darknet way.

[@joooooojo2235]

and by the way, I love your videos and they really help a lot!!!! just a recommendation, it would be to see you use it on more real-world application!!!

[@TheAIGuy]

@@joooooojo2235 the performance is pretty similar between the two, just that darknet is in C, while this allows a python implementation through tensorflow.

[@joooooojo2235]

@@TheAIGuy Thanks for the answer! For the running time is also similar for both of them? I feel like I wasted a lot of time to install darknet while there is another good option :(

[@muslam555]

What are the two programs called in the taskbar "C", (Green) & "C", (Red)?

[@solomonnimrud9627]

hi, i got this error when using "python load_weights.py", could you help me F tensorflow/stream_executor/lib/statusor.cc:34] Attempting to fetch value instead of handling error Internal: failed to get device attribute 13 for device 0: CUDA_ERROR_UNKNOWN: unknown error

[@bansikhubchandani9743]

A very helpful video. Could you please make a video on implementation of a yolo network from scratch??

[@RobTrahms]

Excellent video and repo! Any plans to create a YoloV3-Tiny model?

[@TheAIGuy]

Yes you can use the video and repo from my video called “Real-time Yolov3 on webcam and video” to see how!

[@benson827]

you are an absolute legend

[@jinsam552]

Hi your tutorial is so great. I want to ask if there are some ways to count objects from the video.

[@mellarksgirl]

Yes plsss tutorial on this

[@anayaggarwal6877]

This is my error: ModuleNotFoundError: no module named 'cv2'. It happens when I am running the file through the program. Also, when I was installing the dependencies it said could not find a version that satisfies the requirement tensorflow==2.0

[@anayaggarwal6877]

Wait I fixed it, now when I run it it doesn't do anything

[@zohaibjan]

Can you make a video on TensorFlow please ?, I can reuse code but would like to be able to write my own as well.

[@tr6x679]

How can I use the python detect.py... command with the built-on camera on the Jetson TX2?

[@sirsmacko7444]

Is there a way to utilize the gpu cause my cpu is being pushed to its limits

[@TheAIGuy]

checkout my latest video, it has all the same functionality of this video and more. It will show how to run the detections utilizing GPU and get way faster results.

[@sirsmacko7444]

Thanks

[@rahuldass4882]

Another great one! I'm curious if you found any improvements in performance when using/training YOLO v3 with TensorFlow as opposed to without, especially when doing real-time classification? Also, I was wondering if you had any plans for building a YOLO v3 model with PyTorch as well? It would be interesting to compare using both libraries to assess model accuracy and/or performance with still images and real-time video data.

[@shivangshekhar3552]

I personally faced an issue where the accuracy for the tensorflow detection was not as good as when I was using darknet for the detection... Anything I might be missing out?

[@jackgaul5144]

My loading weights is taking forever. I am not getting an error but I basically freezes my computer and I have to restart.

[@harrydobbs4179]

Hi there, Great video. I have subscribed and liked! Just wondering in the COCO data-set it does not claim it can detect plates, only bowls, forks and knifes. Could you please confirm you are able to detect plates? Cheers!

[@erolgyte]

when i run the load_weights.py, i got this error. ValueError: cannot reshape array of size 258012 into shape (256,128,3,3)

[@Mrduirk]

quiero hacer un proyecto robot con SparkFun Edge , y tflite, a ver si voy encaminado ....yo le meteria un object detection que detectara un objeto y luego un entreno con tensorflow con un aprendizaje profundo y tf agent para arduino que aprenda a cojer el objeto lo que esta viendo con el object detection...van por ahi los tiros??gracias!

[@pratikkorat790]

You nailed it...🤩😁🔥🔥🔥🔥

[@richardjoy8925]

what changes should i make to have a live video detection, thank you!!

[@TheAIGuy]

Richard Joy like from a webcam?

[@richardjoy8925]

@@TheAIGuy yes exactly

[@TheAIGuy]

Richard Joy I’ll update the code to allow webcam video in the next couple days for you! Subscribe to the channel to be alerted when I do!

[@richardjoy8925]

@@TheAIGuy thankx

[@TheAIGuy]

@@richardjoy8925 I've updated the code to allow for webcam detections! Just use 'webcam' as the first parameter, refer to the readme example to see exactly how. Hope this helps.

[@davitshermadini7775]

waiting for new video👍

[@TheAIGuy]

Davit Shermadini new video is up! Check it out!

[@pyroswolf8203]

How can ı change detection labels,writing colors or size ?

[@007jsteeve1]

which command line did you uesd

[@TheAIGuy]

I use Git bash for windows.

[@elaborateexpressions271]

This is interesting!

[@EricZhu]

does this work on mac

[@FarzanMNoori]

perfectly working on CPU but not working on GPU. getting these errors tensorflow.python.framework.errors_impl.UnknownError: 2 root error(s) found. (0) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above. [[node yolo_v3_model/conv2d/Conv2D (defined at C:\ProgramData\Anaconda3\lib\site-packages\tensorflow_core\python\framework\ops.py:1751) ]] [[yolo_v3_model/Reshape_68/_973]] (1) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above. any solution?

[@TheAIGuy]

This is usually due to the tensorflow gpu version, what is your tensorflow version?

[@FarzanMNoori]

@@TheAIGuy Thanks alot for the prompt response. I am usingTF 2.0.0 Initially, I installed TF2.1.0 but getting some issues so downgraded it to 2.0.0

[@seeman4171]

Will this work without GPU?

[@TheAIGuy]

yes this will work without GPU as well!

[@elaborateexpressions271]

I just Subscribed to your channel!!!

[@TheAIGuy]

Thanks for the support!

[@CoyaHarris]

Elaborate Expressions hey hun! I’m willing to support you please support back . Let’s stay connected ❤️

[@CoyaHarris]

Elaborate Expressions just subscribe to your channel ❤️

[@matkoschlebikom3630]

hello i need work witch coordinates object , how i extract coordinates from yolo_v3.py

[@sharanrabinsonmohanraj6230]

How to extract image from detected object in tensorflow object_detection(webcam)

[@gabrieliux25]

0:42 nice dog :D (0.25 speed) child is a dog.

[@lorenzoleongutierrez7927]

Great!, thanks a lot!

[@erolgyte]

is there any changes on detect.py?

[@eluquencyfilmscomedyefcome2059]

Awesome

[@willr6823]

hey, good stuff. would love a kubernetes one!!

[@tudoropran1967]

Good vid

[@TheAIGuy]

tudor opran Thanks so much!

[@sociallife5482]

perfect

[@sorpiseth6917]

hi everyone! I am just start study detection with Tensorflow, so I want to ask you how about tutorial digit detection.

[@amateurgamer3186]

Interesting

[@mere2752]

Done

[@darkcast345]

"How to show a github repo and not how code actually works" If you want to grow the channel write the code and explain it. If you want an example do what sentdex does. Also i just found that you just took it straight from the yolo website

[@ravirajsinhdabhi9255]

PS C:\Users\user\yolo-v3> python detect.py images 0.5 0.5 data/images/img_2.jpeg Traceback (most recent call last): File "detect.py", line 146, in main(sys.argv[1], float(sys.argv[2]), float(sys.argv[3]), sys.argv[4:]) File "detect.py", line 40, in main batch = load_images(input_names, model_size=_MODEL_SIZE) File "C:\Users\user\yolo-v3\utils.py", line 25, in load_images img = np.expand_dims(img[:, :, :3]) IndexError: too many indices for array stucked in one place!!!

[@cursedgrape1247]

Image annotation interface(for object detection) check github.com/emadboctorx/labelpix

[@viswatejvarma7792]

teach us programming rather than this clone of project