CSE 455 Assignment 3 Autumn 2012

Skyler Peterson

Welcome to my artifact page for assignment 3 which involves 3D reconstruction of image sets using various Photometric Stereo techniques. In the remainder of this page you will find images of my results for six objects. These images represent the entire given example sets for this project. Each set contains nine images. Two of which represent the normal found at each pixel of the object in the image; one of which is in color format and the other is in a vector format. One of the images is the albedo calculated for the image. Finally, there are three sets of two images taken of the 3D reconstruction. Each set is of the same image, one with and one without the albedo applied to the mesh. Of the three sets, one is a straight on view and, other than the shading, shouldn't appear much different than the albedo map. The second is a profile view and the third is an angled view that I found interesting or revealing in some way.

NOTE: You may click any image in this document to see it in full size.

Click a thumbnail below to skip to that artifact.

In general the assignment appeared to be quite successful. As this page demonstrates, all of the example objects were reconstructed successfully. The "correctness" of these reconstructions can be evaluated in multiple ways; comparison with the physical objects themselves, comparison with the outputs of some accepted "solution" program, and the aesthetic value of the results. The first of these evaluations will be impossible as the physical objects themselves were not provided and for all I know have been lost to mankind. (or at least to me they are) The second of these evaluations will prove the most realistic; as the class was provided a single and uniform solution set to use as comparison. This however may prove subjective as it is dependent on the correctness of the solution itself. The third and final criteria is subject to the eye of the beholder, however this beholder will give some opinions on the subject. I will write about the successful and not-so-successful points following each image set.

Buddah Image Set

Buddah Normals in RGB Format	Buddah Normals in Needle Format	Buddah Albedo Map

Buddah Frontal Depth Map with No Albedo	Buddah Profile Depth Map with No Albedo	Buddah Angled Depth Map with No Albedo

Buddha Frontal Depth Map with Albedo	Buddha Profile Depth Map with Albedo	Buddha Angled Depth Map with Albedo

The Buddha object came out fairly perfect. There were no visible differences between the solution code and mine that I could perceive in any of the albedo, normal, or depth maps. This reconstruction was probably the most perfect recreation as compared to the solution as even the edge lines matched completely.

Cat Image Set

Cat Normals in RGB Format	Cat Normals in Needle Format	Cat Albedo Map

Cat Frontal Depth Map with No Albedo	Cat Profile Depth Map with No Albedo	Cat Angled Depth Map with No Albedo

Cat Frontal Depth Map with Albedo	Cat Profile Depth Map with Albedo	Cat Angled Depth Map with Albedo

As can be seen this reconstruction appears to look good. The surface is smooth everywhere as one might expect a soft cute kitten to appear. The albedo is a bit pixely about the whiskers and the eyes have a glare however this is present in the solution output also. The 3D reconstructions are entirely similar in the inner regions, while the edges are also similar with exception to a few upward facing lines on the left side of the face that do NOT appear in my output. In this sense mine looks better, although very little.

Gray Sphere Image Set

Gray Sphere Normals in RGB Format	Gray Sphere Normals in Needle Format	Gray Sphere Albedo Map

Gray Sphere Frontal Depth Map with No Albedo	Gray Sphere Profile Depth Map with No Albedo	Gray Sphere Angled Depth Map with No Albedo

Gray Sphere Frontal Depth Map with Albedo	Gray Sphere Profile Depth Map with Albedo	Gray Sphere Angled Depth Map with Albedo

This shape is fairly straight forward as it is in fact a sphere. This made it a good shape to test with during the production of the code, since it is simple and I know exactly what it should look like, especially the normals. The normal vector map is symmetric and the albedo looks like the original sphere. As the viewer can quickly verify, the 3D reconstruction does look pretty much like a half sphere and it matches the given output. The only difference I can see between mine and the solution is some outward lines that show up in the solution in the chewed up looking region at the bottom of the profile shots. Mine do not have these outward projecting lines and like the cat makes my reconstruction a little bit better to look at.

Horse Image Set

Horse Normals in RGB Format	Horse Normals in Needle Format	Horse Albedo Map

Horse Frontal Depth Map with No Albedo	Horse Profile Depth Map with No Albedo	Horse Angled Depth Map with No Albedo

Horse Frontal Depth Map with Albedo	Horse Profile Depth Map with Albedo	Horse Angled Depth Map with Albedo

I thought this was one of the more advanced shapes based off; the fine ridges in its skin, the bend in its neck and legs, and the thin legs/tail projecting from the body (not found in any of the other objects). I was pleased to find that the reconstruction was successful and the only differences were found in the edges. Unlike my previous reconstructions (see cat and gray sphere) there are outward lines that can be seen at the top of the head. These lines do show up in the solution code but in a different location (the bottom of the head).

Owl Image Set

Owl Normals in RGB Format	Owl Normals in Needle Format	Owl Albedo Map

Owl Frontal Depth Map with No Albedo	Owl Profile Depth Map with No Albedo	Owl Angled Depth Map with No Albedo

Owl Frontal Depth Map with Albedo	Owl Profile Depth Map with Albedo	Owl Angled Depth Map with Albedo

This reconstruction is similar to the cat one in nature and my comments are similar. This one however, is my least favorite object as its beak blocked the side of its face so he or she kind of looks messed up and for some reason its right eye is flaring out but this shows up in the solution also. Again, my image has a difference in where the outward lines show up. This time I have lines on his or her or its genetically unspecific left ear.

Rock Image Set

Rock Normals in RGB Format	Rock Normals in Needle Format	Rock Albedo Map

Rock Frontal Depth Map with No Albedo	Rock Profile Depth Map with No Albedo	Rock Angled Depth Map with No Albedo

Rock Frontal Depth Map with Albedo	Rock Profile Depth Map with Albedo	Rock Angled Depth Map with Albedo

This lowly rock turned out to be my favorite reconstruction. I liked it for its modest yet complicated form. All of its rough edges, divots, and projections made for a meteor looking 3D object. After applying the albedo the chaos of the form suddenly takes shape as you can easily map the clods and independent rocks to their correct locations. Again, this object was very similar to the solution with exception to the location of the awkwardly outward projecting lines on the edges.