Train detectron2 object detection custom data | Computer vision tutorial

[@ComputerVisionEngineer]

Did you enjoy this video? Try my premium courses! 😃🙌😊 ● Hands-On Computer Vision in the Cloud: Building an AWS-based Real Time Number Plate Recognition System bit.ly/3RXrE1Y ● End-To-End Computer Vision: Build and Deploy a Video Summarization API bit.ly/3tyQX0M ● Computer Vision on Edge: Real Time Number Plate Recognition on an Edge Device bit.ly/4dYodA7 ● Machine Learning Entrepreneur: How to start your entrepreneurial journey as a freelancer and content creator bit.ly/4bFLeaC Learn to create AI-based prototypes in the Computer Vision School! www.computervision.school 😃🚀🎓

[@NeginNajafi-q2g]

your videos are really helpful, thank you so much

[@praveenavala2297]

Thanks for Great Informative video......how to print evaluation models(Accuracy, prediction, f1 score and recall)

[@minhnguyetnguyen1048]

Hi Felipe. Thanks for your great tutorial. Can I ask you a question about my case. I plan to use Detectron2 to train for waste object detection (there are 4 classes of objects). 1/ Can I use detectron2 with the same backbone (retinanet) as your case? 2/ I have annotation in yolo format, do I need to convert to Coco format? 3/ Can I use the same pipeline in your video for my case? Thanks!

[@urumomaos2478]

sos un genio, me salvaste la vida con tus tutoriales. Saludos desde argentina

[@ComputerVisionEngineer]

Me alegro haber ayudado! Saludos!! 😃🙌

[@annerabbit3938]

It's really helpful! Thanks! But what should I do if I want to train on multi-GPUs?

[@neverninetofive]

What is currently the best object detection framework in your opinion? (Faster RCNN, YOLOv8, detectron2, etc.)

[@ComputerVisionEngineer]

In my opinion I think there is not 'best' but each technology has its pros and cons. Choosing the right object detector in a specific project depends on the project's requirements. If unsure what is the best one in your specific use case, you can test them all out and see which one you like the most. 😀🙌

[@omoklamok]

@@ComputerVisionEngineer for me DINO V2 is better

[@shezwad_]

Thanks. Are you able to do the inferencing? AP evaluation

[@sholehuye2737]

Thanks 😁, Is there anything for the number of objects detected from the image, for example the number of buildings in a housing plan

[@abdulrozzaq8132]

Great explanation 🙌🏻

[@ComputerVisionEngineer]

🙂😃

[@resoluation345]

Thank you Phillpe

[@ComputerVisionEngineer]

You are welcome! 😃🙌

[@corentindoche5645]

Hi, thank you for this video. Is the YOLO format possible if we use masks annotation rather than bounding boxes ? Because in an image with a lot of objects to annotate it is way faster with masks. I use Label Studio with Magic Wand tools that allow to just click on the object and it applies a mask on the object, but I can't export in Yolo format. So is it possible to "transform" my JSON format file into a YOLO format ? Or is there any Labelling tools other than Label Studio that provides exportation of masks annotation to yolo format or is it doable only with bounding boxes ?

[@ComputerVisionEngineer]

Do you mean what type of annotations do you need in order to train detectron2 for semantic segmentation? Take a look at detectron2 tutorial, you need to provide the polygons. colab.research.google.com/drive/16jcaJoc6bCFAQ96jDe2HwtXj7BMD_-m5

[@corentindoche5645]

@@ComputerVisionEngineer Thank you for your answer. My main issue now is that with polygons annotation it will take me too much time to label what I want, so I need to find a way to rewrite my raw json file into a COCO or YOLO format if it is possible

[@nikhilbajaj5179]

Hey! I found this video and tutorial very interesting. I wanted to know how can we apply Explainability to the trained model. Maybe also find an metric for comparison of results of explainability??

[@ComputerVisionEngineer]

I will try to do a future video about explainability in machine learning models. 🙌

[@HoangLe-qy4bc]

Hi, when I try to make predictions, my predictions show class ID (numbers) instead of class names (like 'panda','sheep',...). Do you know how to fix this? I use the code exactly like you wrote

[@ComputerVisionEngineer]

Hey, that is ok, making a prediction should return the class id (a number), not the class name. You can turn it into a class name by loading the names from the class.names files as a list and then using the class id as the index of that list. 🙌

[@utkucamldap3242]

Amazing tutorial 👏

[@tomaskurtenperez5080]

Hola Felipe, i train the detector using Yolov8 like your previous tutorial, recolecting more pictures and its working now. My question is: would you recommend Yolov8 or Detectron2? Also, is it good to set the threshold on an number smaller than 0.5? Gracias!!

[@ComputerVisionEngineer]

Hey, glad it is working better now! I would say both frameworks are 'state of the art', it is up to you and your project which one is best for you. I like yolov8 because of easy of use and real time execution time but detectron2 has a very interesting model zoo, so you can try many different pre trained baselines, which is very useful if you are training a very custom model. 🙌

[@ajarivas72]

@@ComputerVisionEngineer I enjoy your videos 🎥 very much.

[@ronitviju5406]

how can i get metrics such as auc-roc score and mAP-50 etc ,is there a script available or built in function from detectron2?

[@abdallamukhaimar7350]

Can you do an episode about how a person can become Computer Version Eng. Thanks :)

[@ComputerVisionEngineer]

Sure! A few months ago I made a video about a computer vision roadmap, which answers the question 'how to become a computer vision engineer?' (here: kzbin.info/www/bejne/epezcqmbpbBojdU) I will try to upload a new video with an update soon. 😀🙌

[@ibrahimkadri2362]

How we can evaluate the trained model and get the Average Precision and so on ? Thank you so much

[@ComputerVisionEngineer]

Take a look at detectron2 getting started collab, there is an example on how to evaluate the model you trained and compute some metrics as the mAP. 🙌

[@brunovolcovinsky1880]

Hola! I'm trying to adapt this example for semantic segmentation. Do you have any suggestion? I already have the YOLO semantic segmentation annotation format. Great videos!

[@ComputerVisionEngineer]

Hey! Take a look at detectron2 getting started Google colab, the example in the notebook is about image segmentation. 💪🙌

[@ajarivas72]

@@ComputerVisionEngineer What frame (yolo,detectron,transformer) to use for estimating the volume of a pile of sugar?

[@MehrsaMashhadi]

Hello. Thank you so much for your helpful tutorial. I just have a question. When I run other algorithms like YOLOv7 or Detr, each epoch takes a lot of time. But here, for 300 iteration, It only takes like 10 mins. What is the reason? Am I doing sth wrong? In your video also you use 6000 for your iteration, and that's a really big number for epochs. Can you explain it to me please?

[@ComputerVisionEngineer]

Hi, the number of iterations in Detectron2 is not the same as the number of epochs. The number of iterations represents the mini-batch processing steps during model training, while the number of epochs in machine learning signifies how many times the entire training dataset is iterated through. I may have called it 'epochs' a few times during the video, but it was a mistake. 😄

[@MihirJaipuria]

how can we resume training from a checkpoint in colab notebook? also how can we see different metrics like AP in colab notebook? I'm using the colab notwbook in this video

[@amesaleh6824]

hey , i am am learning computer vision , the question is how i can draw and access confusion matrix ??

[@insomniacguy1540]

can you please tell us how to get the mAP50 and values like that please ?

[@SwayamMalla]

how do i resume the training from checkpoint???

[@rainz2133]

What version of python? How did you get detectron2 installed? I keep getting errors when trying to install detectron2

[@ComputerVisionEngineer]

I was using Python 3.9 if not mistaken. What errors are you getting?

[@rajivpannuri92]

Hi, Can I use the custom annotations using polygon menthod via labelme?If yes, Can I please ask how?

[@michael3226]

Great video

[@ComputerVisionEngineer]

😃🙌 Glad you enjoyed it!

[@ceritatujuhdesember5393]

Can you share model evaluation for your technique use detectron?

[@SevinçTanış-p2r]

How can I get the confusion matrix? please help me

[@Bit_Maximum]

Thanks!

[@suphotnarapong355]

Thank you

[@ComputerVisionEngineer]

You are welcome!

[@philipgeorgiev]

I cannot find which yolo has annotations for a dataset, there are so many versions.

[@heshamali5208]

could you provide a video for text detection using Swin text spotter?

[@ComputerVisionEngineer]

I will try to. 🙌

[@johnpsmith8132]

Good day, sir. How can I use this code for video object detection? I have a study project to track traffic signs, but my CV lvl is sub zero XD

[@ComputerVisionEngineer]

What objects are you looking to detect? You need to create a dataset and then follow the instructions I provide in the video. 🙌

[@johnpsmith8132]

@@ComputerVisionEngineer I did a classic traffic sign detection. All worked, thank you! But I have one very question: how to make a predicton on CUDA? I installed proper CUDA and torch versions but get error when change CPU to CUDA, it is something about 'torchvision::nms' operation on the CUDA backend.

[@AdithiDeborah]

Should I just change device to 'cuda' while using a GPU? I am getting a CUDNN error. Is there anything else to change?

[@ComputerVisionEngineer]

The default option is cuda, you don't need to specify the device. Make sure your cuda drivers are correctly installed and that pytorch is using the gpu.

[@AdithiDeborah]

@@ComputerVisionEngineer thank you! It works now 😊

[@xinqiliu6761]

Hi, could you kindly share the alpaca dataset you used in the video?

[@ComputerVisionEngineer]

Sure! You can download the dataset following the instructions provided here github.com/computervisioneng/train-yolov8-custom-dataset-step-by-step-guide 💪💪

[@xinqiliu6761]

@@ComputerVisionEngineer Thank you!

[@sethuraman9884]

Sir how can we use this method in lane detection?

[@ComputerVisionEngineer]

The first step is collecting and annotating your custom data, then you can just follow the steps in the video 😃🙌

[@WelcomeToMyLife888]

any videos on segment anything in the pipeline?

[@ComputerVisionEngineer]

Hey, nothing planned but I may use the segment anything model in a future tutorial. 💪

[@WelcomeToMyLife888]

@@ComputerVisionEngineer cool! can't wait to see it. 🥰keep up your awesome work.

[@justhiphop5704]

Is it possible to recognize multiple images in a pdf? Ang extract it to an excel?

[@ComputerVisionEngineer]

Yes, sure, you can use detectron2 to train a model to detect images in a pdf! 💪🙌

[@justhiphop5704]

Wow this is very interesting! Very helpful for the office guys here

[@prafullsharma8052]

@@justhiphop5704 Hi, im doing exactly the same... like to collaborate ?

[@NeuralNetwork-go5zn]

hello, really nice tutorial! I created a custom yolo detection with yolov8m as a base, but when I run it, as soon as the speed of the objects increases, the algorithm loses track of me and starts to "jerk" the video output. Does anyone have an idea how to fix this? (yolov8m algorithm trained on 300 custom images) Thanks so much to anyone who can help me!!

[@ComputerVisionEngineer]

Hey, what do you mean the algorithm starts to "jerk" the video output?

[@NeuralNetwork-go5zn]

@@ComputerVisionEngineer the video output is jerky, not smooth

[@iftekharalammithu5128]

do you have any courses ??

[@ComputerVisionEngineer]

Hey, I have a free course on OpenCV with Python, you can check it out here: kzbin.info/www/bejne/m3Wsm2iCqq5_aqM. Premium courses are in the works. 💪💪

[@connectrRomania]

Didn't know that detectron use same data format as yolo 🤭

[@ComputerVisionEngineer]

Actually, it doesn't. 🤭 But the wrapper I built around it expects the data in yolo format. It is me, I got used to work with that format. So if I am using both algos (Yolo and detectron2) I can just use the same annotations and I don't loose time converting formats. 😅

[@connectrRomania]

smart move, well done is there a way to use the same pretrained weights of yolo inside detectron2 haha

[@ComputerVisionEngineer]

@@connectrRomania haha if I find a way to do it I will let you know 🙌

[@compdetect]

Met this error: File "../Detectron2/train-object-detector-detectron2/util.py", line 121, in get_dicts label, cx, cy, w_, h_ = line.split(' ') ValueError: too many values to unpack (expected 5)

[@merijan-u1y]

Have you resolved it? I have an error. I downloaded my custom dataset in roboflow YoloV8 format. line 125, in get_dicts int((float(cy) - (float(h_) / 2)) * height), ValueError: could not convert string to float: ''

[@prathyyyyy]

Hey there can you share your Linkedin Profile, excited to have conversation with you! have a great weekend.

[@ComputerVisionEngineer]

Hey, I don't use LinkedIn that much, we can connect through discord 😃🙌. Have a great week you too!