- 20 Dec 2022
- 4 Minutes to read
- Updated on 20 Dec 2022
- 4 Minutes to read
Supported Image Files and File Types
- PNG (recommended)
- Pascal VOC (available for Object Detection)
Images Are Resized During Model Training
By default, LandingLens automatically resizes all images to 512x512 during Model Training. This auto-resize does not affect the images on the platform. The platform automatically resizes images to make the training faster and more effective. If you want images to be a specific size for Model Training, resizing your images before uploading them to your Project is recommended.
Maximum Image Size for Model Training
Each Project Type has a maximum image size for Model training:
- Classification: 1600x1600px
- Object Detection: 1600x1600px
- Segmentation: 800x800px
- Anomaly Detection: 512x512px
The maximum image size is measured by the area, not width and height. For example, the maximum image size for Segmentation is 800x800px. To get the area, multiply 800 times 800, so the equation looks like this: 800 x 800 = 640,000. If you want to resize a Segmentation image, the area cannot exceed 640,000 pixels.
- A resize of 1000x600px will work because the area of this resize is 600,000 pixels, which is less than 640,000 pixels.
- A resize of 800x801px will not work because the area of this resize is 640,800 pixels, which is greater than 640,000 pixels.
The table below shows the maximum area (in pixels) for each Project Type:
|Project Type||Maximum Image Size (in pixels)||Maximum Area (in pixels)|
Upload Images to Projects
Upload Images After Model Training
LandingLens will automatically generate Predictions on images uploaded after Model Training. You will be able to see these Predictions in the Data Browser. This feature will save you time from needing to retrain your Model after you upload new images to your dataset.
In the unlikely event that the Model is unable to predict an image successfully, the status Unpredicted will display on the image.
What if an image does not display any Predictions and has the status "Unpredicted"? If this occurs, the Model did not find anything to detect.
- If you have not already trained your Model, or your Model did not successfully complete its training, the Model will not make any Predictions.
- It is recommended to label uploaded images, even if the Model's Predictions are accurate.
Limitations for Image Uploads
- LandingLens does not have any size limitations for image uploads.
- You can upload as many images as you need to your Project.
Number of Images Required for a Project
Upload Pre-Labeled Images
You can upload images that have been labeled in an external application. These pre-labeled images must be in Pascal VOC file format. Pascal VOC (Visual Object Classes) is a file format that includes the label details of its paired image. It essentially tells the platform where a label is on the associated image.
Before you upload pre-labeled images, you must unzip the file; LandingLens will not accept zipped files.
LandingLens will detect the Pascal VOC file type and display it on upload.
The pre-labeled images will display in the Data Browser.
LandingLens simplifies folder structures and flattens all files. Let's say you have one folder (Folder A) and two subfolders (Folder B and C). If you upload Folder A, LandingLens will upload all images from Folders A, B, and C, as if they came from the same folder.
However, if you upload Classified Media to a Classification Project, LandingLens will ignore any subfolders. Referring to the example above, if you upload Folder A, only the images in Folder A will be uploaded; the images in Folders B and C will not be uploaded.