Indexing
During the indexing stage, Vidsy scans images and videos for faces. This computationally intensive process leverages multiple models to extract not only the location of each face but also details such as age, emotions, and head orientation (pitch, yaw, roll). The application offers several features to manage the indexing process effectively:
Database
Run - Stop
This functionality allows you to start or stop the indexing process. If the process is halted before all files are scanned, Vidsy will track the last processed file so that the scan can resume from where it left off. Additionally, if new files are added after a full scan, restarting the scan will ensure that the application only processes the new files.
Clear
This command clears the database of any previously collected indexing results. All files generated during the indexing process will be removed.
This action is irreversible. Once the database is cleared, the entire index will need to be rebuilt.
Folders
By default, Vidsy scans the Pictures folder for files. If your files are located elsewhere, this command allows you to specify additional folders. When activated, a popup will display a list of folders. You can add new folders by clicking on the + sign and remove them by clicking on the trash icon next to each folder.
Process Configuration
Auto Run
This toggle, when enabled, will make the application automatically scan for new files upon startup.
CPU & GPU
These fields allow you to define the number of files processed simultaneously and the type of processor used. The default settings are:
- CPU: Utilizes all but one of your machine's cores.
- GPU: Utilizes all available GPUs on your system.
It's generally not advisable to set the GPU count above one, as this can cause data-transfer overload and restart all GPU processes. However, if you have a high-performance machine, you may experiment with higher counts at the risk of potential slowdowns.
Model Configuration
Model
This dropdown menu lets you choose the model used for processing images. The available models are:
- Small: Faster but less accurate.
- Large: Slower but more accurate.
Important considerations:
- You cannot switch models while the indexing process is running.
- If the system already contains indexed faces, switching models may trigger a recalculation of some or all indexes. This is necessary because indexes can only be compared if they originate from the same model, requiring all faces in the index to be processed by the same model.
Equalize, Mesh & Accuracy
In the default configuration, a mesh is calculated for each face. This not only determines face and eye orientation but also verifies whether the face is human. This can sometimes result in missed faces. To optimize the algorithm based on your needs, you can adjust the following settings:
- Equalize: When enabled, the image is equalized before processing, potentially improving match quality in high-contrast images.
- Mesh: When disabled, no mesh is calculated. This may result in more faces being found but also increases the likelihood of false positives.
- Accuracy:
When mesh is off, set accuracy higher to reduce invalid faces.
When mesh is on, this can be set lower to capture as many faces as possible.
- Small model: Recommended around 70% accuracy.
- Large model: Recommended 90% or higher accuracy.
By fine-tuning these settings, you can achieve a balance between speed and accuracy tailored to your specific needs.