From the past week, I could not get appropriate data from trained model. Therefore, let’s start again from the beginning of the tutorials. I saw the videos below, and started to train model following the instructions. Custom Object Detection Overview– Collect Images (more than 1000)– Annotate images-Train I download 300 images using Google Image Crawler, and deleted odd images or images with multiple carrots. I tried to train with this data, however, there comes out for an error. The error is about shape, so I deleted all png files on the list. In the past, deleting every png files make … 2019.09.30 How to train YOLO with my own dataset? – 4 : Failure again 더보기

Today, I want to check whether it is available for operating two webcam in one program. And I suceeded to run two web cameras in one program. Reference : https://stackoverflow.com/questions/29664399/capturing-video-from-two-cameras-in-opencv-at-once My next step is to run yolo with two cameras 2019.09.27 How to run two webcams in one program? 더보기

The loss of 1000 epoches with 5 images are around 2. I tested with webcam whether yolo could recognize carrot or not, however, the result is as same as 100 epoches. Yolo could not recognize carrot. Therefore, I retrained yolo with more datasets – 11 with 1000 epoches. After testing with this model and if this model does not work properly, I will collect real data or real pictures with my own dataset. After training 1000 epochs, results are still unavailable to detect carrots. I have to find the reason for this error. 2019.09.27 How to train YOLO with my own dataset? – 3 : what is the result of the training 1000 epoches of YOLO? 더보기

From the last training, the loss of the model changes 12 to 7. I tested the model with webcam, however, the accuracy is lower than I expected. As in the picture, yolo started to recognize other objects as a carrot, so I have to train with clear data. Since I used the pictures with many carrots, I have to select the picture with only one carrot. Starting from small dataset, I have to increase numbers of data and epoches – 1000. If the loss decreases around 0, then I expected the yolo could detect carrots as well. After 1000 epoches, … 2019.09.26 How to train YOLO with my own dataset? – 2 더보기

Now, yolov2 could not recognize carrots as well as person. Then I want to train yolov2 cfg files with my own datasets of carrots. I crawled 200 carrot images from Google images using image crawling as I previously posted. Then I used darkflow to train yolo models. First, I installed labelImg from github in the Anaconda3. Then I manually select the carrot box and saved it to the .xml files. Then I followed the blog below, and tried to train yolo model. However, there is a problem with error message : AttributeError:’Nontype’ object has no attribute ‘shape’. I found the … 2019.09.25 How to train YOLO with my own dataset? – 1 더보기

How to cut acryl with laser cutter? To make a cover of Jetson TX2, first draw basic drawings using Rhino 3D. Then, export files as a dwf file. With the laser cutter below put acryl in the laser cutting machine and push the start button. Today, we use the material which is not an acryl , so there was a failure while cutting the material. However, at 5th floor with the acryl plate I could finally get the completed results of the acryl cover for Jetson. Below is the result of today’s work. 2019.09.25 How to cut Acryl with laser cutter? 더보기