Represent a BGR image.
Crop a cropWidth
x cropHeight
patch from an image.
Normalize a BGR image.
Normalize a BGR image. The normalize is per channel. Each pixel will minus mean value of the channel. Then divide std value of the channel.
Each pixel value of the input BGR Image sub the given mean value of the corresponding chanel
Random crop a specified area from the Image.
Random crop a specified area from the Image. The result is also an image
Convert a batch of labeled BGR images into a Mini-batch.
Convert a batch of labeled BGR images into a Mini-batch.
Notice: The totalBatch means a total batch size. In distributed environment, the batch should be divided by total core number
Convert a BGR image to dense vector of spark mllib
Write a BGR image sequence into one or many hadoop sequence files.
transform labeled bgr image to sample
Convert a byte record to BGR image.
Convert a byte record to BGR image. The format is, first 4 bytes is width, the next 4 bytes is height, and the last is pixels coming with BGR order.
Convert byte records into grey image.
Process an image with brightness, contrast, saturation in a random order
Image crop method, e.g.
Image crop method, e.g. random, center
Represent a grey image
Crop an area from a grey image.
Crop an area from a grey image. The crop area width and height must be smaller than grey image width and height. The area position is random.
Normalize a grey image.
Normalize a grey image. Each pixel will minus mean and then divide std.
Convert a batch of labeled grey images into a Mini-batch.
Convert a batch of labeled grey images into a Mini-batch.
Notice: The totalBatch means a total batch size. In distributed environment, the batch should be divided by total core number
transform labeled grey image to sample
Flip a image with a probability.
Flip a image with a probability. The threshold higher, the less easier to flip the image.
A BGR Image with label.
A grey image with a float label
Lighting noise for data augmentation
Lighting noise for data augmentation
Krizhevsky et al. proposed fancy PCA when training the famous Alex-Net in 2012 Fancy PCA alters the intensities of the RGB channels in training images For each training image, add the corresponding quantity to each RGB image pixel
Represent a local file path of a image file with a float label
Read BGR images from local given paths.
Read BGR images from local given paths. After read the image, it will resize the images to the given width and height. Besides, it will also divide the pixel value by the given normalize value.
Read BGR images from local given paths.
Read BGR images from local given paths. After read the image, it will resize the images to the given width and height. It will also divide the pixel value by the given normalize value. Besides, it will return the file name.
Read BGR images from local given paths.
Read BGR images from local given paths. After read the image, it will resize the shorted edge to the given scale to value and resize the other edge properly. It will also divide the pixel value by the given normalize value.
Read BGR images from given paths.
Read BGR images from given paths. After read the image, it will resize the shorted edge to the given scale to value and resize the other edge properly. It will also divide the pixel value by the given normalize value. Besides, it will return the file name.
Read byte records from local hadoop sequence files.
Multi-thread convert BGR images into Mini-Batch.
Multi-thread convert BGR images into Mini-Batch.
Note: the source iterator must be thread safe.
crop the center of image
crop the random position of image
Read BGR images from given paths, return LabeledBGRImage and the file name.
used for image object detection visualize detected bounding boxes and their scores to image
Crop a
cropWidth
xcropHeight
patch from an image. The patch size should be less than the image size. There're two cropping methods: at random and from the center. The former is preferred for simple data augmentation during training while the later applies to validation or testing