Custom YoloV7 TensorRT Jetson Nano | Yolov7 | Rocket System

so this is the YOLO V7 tiny model converted into 1030 engine working perfectly fine on Jetson Nano hello and welcome to Rocket systems YouTube channel so in our previous video we discussed how we can convert a YOLO V7 pretend model into a tensor RT engine on Jetson Nano now in this video we are going to focus on a custom trade model so I have basically trained a custom YOLO V7 tiny model on this particular video so this is a video of the toys and I've extracted images from this video and then I've trained the YOLO V7 tiny model and this is the model which I have this is the batch.pd file and this is how the results look like so you will notice that this roller is getting detected this car is also being detected and then this phone is also being detected perfectly fine so in this video what we are going to do is that we are basically going to convert this custom YOLO V7 tiny model into a tensority engine and then we will run this on the Jetson Nano to see how it performs so let's get started so before we start this video I just want to give a big shout out to one of the guys I was discussing on how you can convert YOLO V7 into tensor RT I actually tried converting yellow V7 tiny into tensor RT a lot of time but I was having a few issues with the boundary box and some other issues so this gentleman on my last video actually commented and suggested that to try out the config files from the deploy directory so when I retrained it and converted it back it worked perfectly fine so what basically you need to make sure is that when you are trading your YOLO V7 custom model instead of using the so instead of using the files from this directory training directory thank you you need to make sure that you are using the files from the deploy directory so what I did I actually used this particular yml file in my training and then and then when I converted the model into denser RT it worked perfectly fine on the Jetson Nano now let's get on with the video okay so what I've done is that I have copy pasted my where is the file oh okay so yeah so I've copy pasted this bench dot PT file this is the custom YOLO V7 tiny model and then now let's start the conversion process and we'll convert this into the tensor RT engine now I will simply again open the build step.hd file and again if if you don't know already so this is the the Jetson YOLO V7 tensor RT repository you can clone this repository the link is given in the description and you can you know follow this repository to convert your YOLO V7 models pretend or custom model into the tensor RT engine so I'm just going to run the First Command which will help me convert my PT file into that WDS file so I'm just going to name it as W best DOT WTS okay so if we'll now look here so our batch.wts file is now generated here now next thing what we need to do is that we need to move inside the YOLO V7 directory and then here we will create a directory build then we will move inside the build directory and then we will copy our best or WTS file here now before we can run the cmake or make Command we have to update the config file and here because we are not using the pretend model so we are using our custom models so I will rename this to three and then let's go to the terminal and here let me just close this file so here we will simply run cmake [Music] foreign and now we will simply run make Command okay so the make Command is also complete now we can simply run this command to generate our tensority engine file so let me just clear this sudo dot slash YOLO V7 minus s patch Dot WTS best DOT engine and then I'm just going to use T because it's a tiny model and let's wait for it to build it successfully perfect so the engine file is now successfully built so at this point if you want we can directly test this using the images files but we don't have the images for our custom model so what we are going to do is we are going to directly test this on the video file so I have also pasted here the toys dot MP4 which I trained for and let's open the app.file and let's rename let's rename here and let's mention the toys dot MP4 let's close this now because this is the custom model so what we need to do is we also need to update the YOLO detection.pi file so let me just open this and I'll explain you so here you will notice that we have self dot categories so we need to comment this out [Music] because these categories are from the pre-trained model and we don't have a pre-trained model we have a we have our own custom model so we need to write our classes here so the classes which we have is roller and then oops car and then phone so this is the important step which you need to make sure when you are doing I mean intensity over the custom model please comment out these default categories which are mentioned here and mention your categories which you have now we can simply close this and then we can run this app.file so now let's run our app.pi file okay so I think we also need to okay so we also forgot to rename our engine file so because this we are using the custom engine file so this is not YOLO V7 this is best DOT engine file now let's run it again and you can see our detection is happening perfectly fine and the model is now detecting our roller let's wait for the car to be also detected and the card is also being detected perfectly fine so this confirms that you can run YOLO V7 tiny custom model converted into a tensority engine on Jetson Nano and this seems to be working perfectly fine so I guess that's all for this video thank you for watching this video please like share and subscribe to the channel [Music] thank you [Music]

Channel: Rocket Systems

Views: 3,653

Keywords: Custom YoloV7 TensorRT Jetson Nano, YoloV7 TensorRT Jetson Nano, TensorRT Jetson Nano, Yolov7 Jetson Nano, Custom YoloV7 Jetson Nano, jetson nano, custom yolov5 tensorrt engine on jetson nano, running custom yolov5 tensorrt engine on jetson nano, yolov7, yolov7 custom dataset, nvidia jetson nano, como configurar jetson nano, tensorrt on jetson nano, nvidia jetson nano developer kit, tensorrt jetson nano, yolov5 tensorrt engine on jetson nano, object detection jetson nano

Id: n9BSrfqpVFA

Length: 8min 11sec (491 seconds)

Published: Mon May 01 2023