The following zip files contain same-sized image 'chips' of various object categories. The data was extracted from the LabelMe dataset. There are two files per chips; namely, chipX.jpg and chipX.mat. chipX.jpg is simply the object cropped form the image it came from, and is provided for quick previewing. The Matlab mat chipX.mat file contains some meta information (like the chip's source image in LabelMe) and its response to a bank of filters (pass-through, laplacian, x/y-derivatives, etc...). The filtering was done before the chips were cropped out to avoid bordering artifacts. Back to the main page.

Here is that Matlab code that generated the contents of the zip files. It requires the LabelMe API.

Frontal screens (338), Bookshelves (31), Keyboards (162), Side-view cars (278), Frontal/rear-view cars (309), Bottles (71), Stop signs (41), Fire hydrants (67)

Example. Frontal Screens.


Example. Side-view cars.


Example. Frontal/rear-view cars.


Example. Bottles.


Example. Stop signs.


Example. Fire hydrants. The hyrdant annotations are inconsistent in LabelMe (bounding boxes too loose, or too cropped).