Projects 2009

From Computer Vision, CIS 581
Jump to: navigation, search

Project 1: Canny Edge Detection

Test Images

I1.jpg I2.jpg I3.jpg

Test Image Truth

Berkeley Segmentation Dataset 189090

I2_truth.jpg

Berkeley Segmentation Dataset 189090

I3_truth.jpg

Team Results

Team 1

T1_I1.png T1_I2.png T1_I3.png


Team 2

T2_I1.png T2_I2.png T2_I3.png


Team 2 (Revised)

T2b_I1.png T2b_I2.png T2b_I3.png



Team 3

T3_I1.png T3_I2.png T3_I3.png


Team 4

T4_I1.png T4_I2.png T4_I3.png


Team 4 (Revised)

T4b_I1.png T4b_I2.png T4b_I3.png



Team 5

T5_I1.png T5_I2.png T5_I3.png


Team 5 (Revised)

T5b_I1.png T5b_I2.png T5b_I3.png


Team 6

T6_I1.png T6_I2.png T6_I3.png


Team 7

T7_I1.png T7_I2.png T7_I3.png


Team 8

T8_I1.png T8_I2.png T8_I3.png


Team 9

T9_I1.png T9_I2.png T9_I3.png


Team 10

T10_I1.png T10_I2.png T10_I3.png


Team 10 (Revised)

T10b_I1.png T10b_I2.png T10b_I3.png


Team 11

T11_I1.png T11_I2.png T11_I3.png


Team 12

T12_I1.png T12_I2.png T12_I3.png


Team 12 (Revised)

T12b_I1.png T12b_I2.png T12b_I3.png


Team 13

T13_I1.png T13_I2.png T13_I3.png


Team 14

T14_I1.png T14_I2.png T14_I3.png


Team 14 (Revised)

T14b_I1.png T14b_I2.png T14b_I3.png

Project 2

Test Image

project2_testimg.png

Test Methodology

The goal of this evaluation is to choose the simplest possible image that will demonstrate whether the triangulation based morphing and thin plate spling morphing are operating correctly. Starting from the reference test image, team code is used to warp the image into four source images such that the center point is translated along four cardinal directions. Team code is then used to smoothly morph between these four keyframes, using only five control points: four in the image corners and one central control point at the intersection of four triangles. A matlab script is available to replicate this test.

Team results

Triangulation Evaluation

<youtube>5ZL_P5Os1PE</youtube>

Thin Plate Spline Evaluation

<youtube>9U3PFvFdMUQ</youtube>

Project 3

Test Images

project3_testimage_1.jpg project3_testimage_2.jpg

Test Methodology

The two test images contain 9 unique fiducial markers, which are on a plane parallel to the first image. These fiducials are visible in both images, however the second image undergoes a translation and a small rotation. The fiducials are planar, so the transformation between fiducials is expressible as a homography. A correct mosaic will align the fiducials. These test images were chosen to stress the adaptive non-maxima suppression for the Harris corner detection (e.g. multiple corners per fiducial), the blurred feature descriptor (rotated image), RANSAC (multiple possible correspondences based on which fiducial corner is chosen) and the blending (red vs. blue).

Also shown are the mosaic results generated by the team to best showcase their approach.

Team results


Team 1

T1.png T1_mosaic.jpg


Team 2

T2.jpg T2_mosaic.jpg


Team 3


T3.png T3_mosaic.jpg


Team 4

T4.png T4_mosaic.jpg


Team 5

T5.png T5_mosaic.jpg


Team 6

T6.jpg T6_mosaic.jpg


Team 7

T7.png T7_mosaic.jpg


Team 8

T8.jpg T8_mosaic.jpg


Team 9

T9.png T9_mosaic.jpg


Team 10

T10.png T10_mosaic.jpg


Team 11

T11.png T11_mosaic.jpg


Team 12

T12.png T12_mosaic.jpg


Team 13


T13.png T13_mosaic.jpg


Team 14

T14.png T14_mosaic.jpg

Final project

  • Project 4: due December 15
    • Milestone, November 30
    • Project presentations, Tuesday December 15, 12-3pm in Towne 303