Image quality check
Opens every image and measures its size, sharpness, exposure and contrast; flags broken, small, blurred, dark, bright or flat images with the reason.
How it works
Each image is measured at the same scale: sharpness is the variance of the Laplacian (low = blurred), brightness and contrast come from the grey levels, and the share of nearly black or nearly white pixels shows bad exposure.
Sharpness is judged against the images of the same folder: an image under Blurred (%) of its folder's typical sharpness is flagged blurred, so sharp-edged and soft-textured classes are each judged fairly. Nothing is removed here: the table goes to Curate and split (which removes the flagged images) or to a Table and a Chart. Runs on any worker, a few hundred images per second.
Inputs
| Input | Takes | What it is |
|---|---|---|
Imagesimages | Files jpg, jpeg, png, webp, bmp, tif | Images, in folders (the folder is the label) Takes several connections. |
Outputs
| Output | Gives | What it is |
|---|---|---|
Flagstable | Table csv | quality.csv: one row per image (path, label, width, height, sharpness, brightness, contrast, dark_pct, bright_pct, flag, reason); flag is empty when the image is fine |
Live telemetrytelemetry | Live telemetry | What the block sends while it runs, for views such as Chart or 3D view. |
Settings
| Setting | Default | What it does |
|---|---|---|
Smallest side (px)min_side |
0 | Images smaller than this on their shorter side are flagged; 0: no limit. Many datasets have small photos (FruitNet: often under 224 px): set it to what your model needs From 0 |
Blurred (%)blur |
25 | Flagged under this share of the folder's typical sharpness; 0: no check From 0 to 100 |
Too dark (% black pixels)dark |
60 | 0: no check From 0 to 100 |
Too bright (% white pixels)bright |
60 | 0: no check From 0 to 100 |
Least contrastcontrast |
8 | Standard deviation of the grey levels; 0: no check From 0 |
Where it runs
On one of your computers with the EsaSim worker: Any worker.
The worker installs the Python packages it needs when they are missing: numpy</code>, <code>pillow.
In its code, settings are ctx.params["key"], inputs ctx.inputs["id"] and outputs ctx.output("id", "name.ext"); see Custom blocks.