A console-based wrapper to Alturos.Yolo v3.0.6. Counts the number of each type of object found within each image.
The compiled binary can be downloaded from https://github.com/ryanboyd/Igor/releases
Run from the command line with two parameters: the folder that contains your input files, and the location where you would like to save your CSV file. For example:
igor.exe "C:\DataFolder\Images" "C:\DataFolder\Output.csv"
Igor will recursively search your input folder for all jpg and png files.
Alternatively, you can run igorGUI.exe and perform the same analyses from a graphical interface. The underlying core does all of the same processing, but this version might be slightly more accessible to people who aren't as savvy/comfortable with command lines.
Finally, when you download Igor, you may notice that there is a "modelDat" folder included in the archive. Within the modelDat folder, you will find several subfolders containing pre-trained models, plus a "modelSelect.txt" file. For the console-based version of Igor, you can change the details in your modelSelect.txt file to tell the software which pretrained model to use. In the GUI-based version of the software, you can use the interface/buttons to select your model files. Note that you will want to make sure that you're selecting all three files from the same model folder.
By default, Igor processes images using your CPU. If you want Igor to use your GPU instead (which is much faster), you will need to do a little bit of extra setup by following these instructions from the original Alturos.Yolo repo:
- Install the latest Nvidia driver for your graphic device
- Install Nvidia CUDA Toolkit 10.2 (must be installed with correct hardware driver for CUDA support)
- Download Nvidia cuDNN v7.6.5 for CUDA 10.2
- Copy the
cudnn64_7.dllinto the same folder as igor.exe
This assumes, of course, that you have a compatible GPU in the first place.