Reconstruction of Blood Transfer Stains

Photograph the entire blood stain on a surface with a uniform white (or very light) background. Crop tightly around the blood stain — but keep the entire stain fully visible without cutting off any edges. Note the swipe direction, which our app will reconstruct.

For best results:

• Use a clean, uniform white background
• Resolution at least 500x200 pixels (higher is better)
• Avoid cluttered backgrounds and crops that cut off any part of the swipe

Examples (click to view full size):

Good example: good_image_4904.png — uniform white background, full swipe visible, resolution > 500x200 px

Bad example: bad_image_4912.png — non-uniform / cluttered background, crop too tight

These open in a new tab so they don't interrupt your current view.

Use the Practice photos strip on the classify screen, just under the upload box (it appears once you are signed in). These are the eleven workshop teaching images, ordered easiest first — work left to right, each one chosen to show you something the one before it could not. Tap a thumbnail and it is loaded exactly as if you had uploaded it yourself, including the step where you outline the stain. Nothing to download, and no third-party account. Once a photo is being worked on the strip tucks itself away to keep the screen short — the one-line Show practice photos button brings it back.

Or make your own: swipe a bloody object along white cardstock and photograph it — see "How do I photograph a swipe?" below.

Yes - it is the biggest improvement you control, and it costs nothing. After you choose a photograph the app shows it again and lets you loop, box or stroke the stain before any credit is spent. If you just press Classify, the app segments on its own, which is measurably less reliable about direction.

Threshold (preview) shows you exactly what would be segmented, free of charge, so you can correct your marks before committing. Whatever the answers say, look at that thresholded image first - most surprising results are explained by what was segmented, not by the model.

Every classification is screened for whether the segmented region looks like blood at all. When it is flagged, the object, direction and pattern answers are hidden behind a click - they were computed and are on your record, but showing them under that warning would undercut it. The thresholded stain image stays visible, because that is the evidence you are being asked to check.

If you did not mark the stain, your credit is refunded: nobody told that run where to look, so a miss is ours. A run you aimed yourself is charged either way. Use the "There IS a stain" button to go back and draw round it.

The check also flags about 3 in 100 genuine stains, so it is a prompt to look, not a verdict - and a pass is not a guarantee either.