YOLOv7 Segmentation | Concrete Crack Detection | Google Colab | step-by-step Tutorial

Traditionally, YOLO models were designed exclusively for object detection. However, it has proven to be very influential in creating high-speed image segmentation architectures such as YOLACT. The recently released YOLOv7 natively supports not only object detection, but also pose estimation and image segmentation.

Chapters:

0:00 Introduction
0:57 Roboflow Notebooks
1:47 Installing YOLOv7
2:48 YOLOv7 repository structure
4:28 Inference using model pre-trained on COCO dataset
5:41 Speed vs accuracy trade-off
7:30 YOLOv7 dataset structure
9:32 Download dataset from Roboflow Universe
11:21 Training YOLOv7 segmentation model on custom dataset
11:40 Model evaluation and prediction
13:34 Outro

Resources:

[vikashkumar-cree]

Dear Tutor,
Greetings! I downloaded the Yolo format of the dataset. In the train folder, I can see images and corresponding labels folder, But I can't see any the yellow label that is created on crack. Infact, this is shown in the tutorial, but in the actual dataset, it is missing. It seems like it is not needed as a label is already done.

[robosergTV]

Thanks, great vid!
YOLO7 creates an image with predictions during training, as you've shown it. Is it a feature of YOLO7 codebase, since YOLO5 does not generate an image with predictions automatically for me, only YOLO7 codebase does (as in your video)

[kushshrestha]

This guy seriously needs a pay rise. Awesome content

[christian.js.]

Thanks for this tutorial, I've been searching for hours on how to display/visualize feature map in YOLOv7, please make a tutorial about that. 😁

[angelospapadopoulos]

is it possible and how can we train from scratch here?

[ksteven]

Thank you for your tutorial. I have a question for you.
At around 2:55, you mentioned that you had previously completed a project that involved object detection and instance segmentation at the same time. Would it be possible for me to take a look at the code for that project?
Thx

[joshuamacasadia]

How can i deploy the model to roboflow?

[janithaprathapa]

how to deploy this model in local machine

[dinanmutamaddin]

how if i want to change thickness bonding box and size of font when predict object

[vishalpahuja]

Hi can you show annotation process for cracks on a single wall and detecting.
Thank you.

[abangfikri]

thank u for the video, it helped me alot. However is yolov5 or yolov7 segmentation model can be deployed on android? is it possible? and how can it be done?

[afiedoh]

Hello,
Thank you for your video.
Please how do i apply my best.pt file to realtime video from my webcam?
Thank you

[serchengxu]

can you do a real time crack detection with using yolov7 ?

[nabiladnan]

when start train:
AttributeError: module 'numpy' has no attribute 'int'.

[Laddu]

Hey. I think it's semantic segmentation not instance segmentation.

[rishabhsheoran]

Hey! can you please help me with the deployment of the yolov7 segmnetation model on triton?
When I hit the triton inference server, I get back following outputs:
name: output
tensor: float32[batch, anchors, Concatoutput_dim_2]

name: onnx::Slice_539
tensor: float32[Transposeonnx::Slice_539_dim_0, 3, Transposeonnx::Slice_539_dim_2, Transposeonnx::Slice_539_dim_3, 40]


name: onnx::Slice_693
tensor: float32[Transposeonnx::Slice_693_dim_0, 3, Transposeonnx::Slice_693_dim_2, Transposeonnx::Slice_693_dim_3, 40]

name: onnx::Slice_844
tensor: float32[Transposeonnx::Slice_844_dim_0, 3, Transposeonnx::Slice_844_dim_2, Transposeonnx::Slice_844_dim_3, 40]

name: 517
tensor: float32[Mul517_dim_0, 32, Mul517_dim_2, Mul517_dim_3]

output <class 'numpy.ndarray'> (1, 100800, 40)
onnx::Slice_539 <class 'numpy.ndarray'> (1, 3, 160, 160, 40)
onnx::Slice_693 <class 'numpy.ndarray'> (1, 3, 80, 80, 40)
onnx::Slice_844 <class 'numpy.ndarray'> (1, 3, 40, 40, 40)

From the above outputs how do i extract the bounding boxes and the masks?