The following pictures present snapshots taken during the rendering process. The location of the photon is indicated by a red circle. I used a clipping plane to eleminate geometery behind the photon. (Essentially limiting the effect of the light to a hemisphere, rather than a complete sphere.) The paper does not mention that this needs to be done, but examining the example photos, it appears that the author is also doing this... Also, the author indicates that the images in the accumulation buffer should be scaled by 1/N, where N is the number of initial photons. I believe this should be 1/M, where M is the number of images.

Technically, there are actually more bounces than shown here. The last photon should continue three more bounces. However, The table was getting pretty unwieldly... I would guess that you get the general idea looking at this incomplete set.

Here is the completed image:

Photon Origin

One Bounce

Two Bounces