The “Where’s Munna” Pics (including spoilers)


Before the spoilers… some general information

  1. GitHub page, where you can find Python, Lua and Adobe ExtendScript code that processed the data and generated the visuals:

  2. Details on the common techniques used in the facial recognition field and our implementations of them in the video. It is intended for an advanced high school or undergraduate level audience. This link also has high quality graphics that can be used for educational purposes, released under a Creative Commons license:

  3. The spoilers for “Where’s Munna,” can be found below. Make sure you try to find him in all the images before peeking!

  4. The data on average human performance can be found below. Or click here to skip to it:


Spoilers for “Where’s Munna”

Click on a picture to find out how other people did.


Data Summary

  1. All pictures were selected so that the our personal laptop (Macbook Pro 2015) was able to find Munna. In other words, the accuracy for the computer was always 100%.

  2. 100 human participants comprised of high-quality workers from MTurk.

  3. IOU threshold threshold was 0.30 in order to be counted as “found.” This is a very generous criteria and we set it low since MTurk presentation was not ideal. The human time reported below was adjusted for the same reason.

  4. For more details on the experiment setup and a detailed, participant-by-participant summary on the task, fill out the request form here.

The Primary Result

Our old, personal laptop was able to solve the “Where’s Munna” pictures 5 times faster than humans. On average, across all 10 pictures, the laptop found Munna in 5.60 seconds. Humans took 25.49 seconds.

Human Accuracy Human Time
Laptop Time
1 86% 25.07 16.68
2 32% 45.19 2.4
3 72% 22.74 9.98
4 96% 17.48 1.17
5 96% 15.25 0.9
6 93% 10.11 6.07
7 49% 35.34 6.09
8 76% 37.10 2.03
9 56% 31.23 10.17
10 96% 15.37 0.64