think.image.patch — thinktopic/think.image 0.4.16

*content-threshold-cutoff* clj

buffered-image-has-alpha-channel? clj

(buffered-image-has-alpha-channel? img)

Return true of this image has alpha channel

content-rect->gray-image clj

(content-rect->gray-image src-img content-rect)

Without a specific method to get a precise content mask we can create one from the content rect returned from image->content-rect.

double-to-long clj

(double-to-long item)

generate-n-filtered-rects clj

(generate-n-filtered-rects rect-filter-fn
                           content-rect
                           rect-count
                           rect-dim
                           &
                           {:keys [max-src-rect-count]
                            :or {max-src-rect-count 500}})

Generate max-src-rect-count random sub-rects, filter by a given filter function. Return rect-count of those.

image->content-rect clj

(image->content-rect img)

image->edge-content-mask clj

(image->edge-content-mask img)

image->patch clj

(image->patch img
              &
              {:keys [rect datatype colorspace data-array normalize]
               :or {datatype :double colorspace :rgb normalize true}})

By default, turns BufferedImage img into an array suitable for cortex input. If :normalize set to false, returns RGB array of pixel values.

image->patch-impl clj macro

(image->patch-impl data-array num-pixels cast-fn array-fn normalize?)

image->patches clj

(image->patches img
                patches-per-image
                patch-dim
                datatype
                &
                {:keys [image-augmentation-fn]})

If an image has an alpha channel, then we assume it has a mask. Else we assume it is safe to create patches out of the entire image.

image-src->patches clj

(image-src->patches img-src
                    patches-per-image
                    patch-dim
                    datatype
                    &
                    {:keys [image-augmentation-fn]})

Given a filename produce a set of random rgb patches of a given datatype potentially augmented.

is-rect-completely-within-mask? clj

(is-rect-completely-within-mask? mask-img rect)

(is-rect-completely-within-mask? mask-img rect byte-data)

Is this rect contained by the mask

masked-image->patches clj

(masked-image->patches img
                       mask-img
                       patch-count
                       patch-dim
                       content-rect
                       datatype
                       &
                       {:keys [image-augmentation-fn]})

patch->image clj

(patch->image data img-width)

patch-mean-subtract clj

(patch-mean-subtract patch r-mean g-mean b-mean & {:keys [bgr-reorder]})

Subtracts means of global images from each channel and optionally reorders RGB->BGR. Use for ResNet image preprocessing.

pixel-byte->double clj

(pixel-byte->double byte-item)

random-sub-rect clj

(random-sub-rect outer-rect output-dim)

Create a square sub-rect inside the outer-rect