Category: Uncategorized

Complaining comes easily for me, but I’d rather be positive and creative than negative. Here’s a strategy I try to employ when I see some art I just don’t agree with.

First, I must be very careful about how I use words like “good”, “bad”, “right”, or “wrong” to describe art. Someone decorating his house in colors I don’t like isn’t wrong. If he likes it, then it’s working as intended. If the artist’s intent is clear, I can point out elements that support or detract from that goal, or I can make a judgement on whether that goal is good or not.  For example, if a director says his movie is about the horror of random violence, but the movie has cool, fun car chases with lots of collateral damage that doesn’t upset the protagonists, he’s failed to make random violence horrifying, and it’s appropriate to use the word “failure”.  On the other hand, if a movie is set in World War One and doesn’t address the themes that I think are important about World War One, that’s not a failure of the movie. That’s a mismatch of expectations. I might make the argument, “It’s irresponsible to represent WWI in this way,” but I can’t say, “The writer forgot this obvious thing.”

I don’t have much to say about art that doesn’t do anything for me.  What really sticks in my mind and bugs me is art that does a lot of things that move me, except for That One Thing.  I’m more likely to pick a tomato slice off a delicious hamburger than I am to try to eat a salad made entirely of ingredients I dislike.

I’ll think something like, “That scene was so emotional, but she should have said this instead.”  How presumptuous to think that I know the character better than the person who plays her every week!  What I see as a mistake is a mismatch between the version of the character the actor knows & expresses through her acting, and the version of the character I’ve constructed in my mind.  My version of the character has gone through three lossy conversions:

1. The actor doesn’t have the opportunity to express all she knows about the character in the scenes in the show.
2. I don’t notice or remember everything the character did
3. The mental model I build based on those actions is strongly colored by my own beliefs and experiences.

Not only do I lack the knowledge required to tell the actress how to play her character, but I don’t have the relationship to start that conversation.  She’s doesn’t even know I exist. So thinking about how to “fix” that “missed opportunity” in whatever art I’m mostly enjoyed is wasted effort.

Instead of complaining about someone else’s art and trying to fix it for him, I draw inspiration from the parts I like, and make a new thing that includes other things I like.  What’s my version of Character X?  What would this setting look like through my philosophical lens?  Understanding what I don’t like about a thing and how I would build it differently forces me to examine and explain my beliefs, which is great for life, not just art.

I throw out ideas that I’ve generated in this way pretty often.  If I’m applying my strategy correctly, they won’t sound like sub-tweets.

Start with Part 1 to learn how I captured 4000 photographs of the mostly-naked, mostly-painted Solstice Cyclists.

Inadequate spreadsheet

My naive start was a spreadsheet with columns for what people wore, what they rode, and a description of their paint.  I used the row number of the spreadsheet as the cyclist ID & tagged images that contained a certain cyclist with that number.  I had columns for top, bottom, head, and face clothing.  That didn’t account for people wearing fairy wings, or sunglasses & a fake beard at the same time. Putting each piece of data in a separate column meant that I could search for “red” or for “sunglasses” but not for “red person wearing sunglasses”  So the spreadsheet was not expressive enough to capture the information & search was inefficient, so dupe-checking took a long time.

Database design

Instead of giving each cyclists clothing slots that could have either 0 or 1 items in each, I created a many-to-many relationship between clothes and cyclists. Each piece of clothing also had a “slot” attribute (top, bottom head, face, back, or other). So a cyclist could wear any number of items, and each item would keep track of where it was worn.  Cyclists & images also had a many-to-many relationship. Vehicle & Sex were simple enumerations.  Descriptions remained as plain text.

Spreadsheet to database.

Converting all the data in the spreadsheet to DB records let me remove any inconsistencies in how I entered the data in the spreadsheet, e.g. “wig, blue” or “blue wig”.  As I added clothes & vehicles to the DB, I searched & replaced those words in the spreadsheet with the DB IDs.  I had to be careful to replace only words in the appropriate columns, since the plain-text descriptions sometimes referenced clothing or vehicles. Sometimes I missed and found a description like, “Mostly red, wearing a green 73” which is quite confusing.

Once I’d replaced all the words with database IDs, I exported the spreadsheet as a CSV file and wrote a PHP script to ingest it into the database. I chose PHP because I’ve already done a lot of SQL with PHP for my Atlanta Fashion Police & convention gallery projects.  The script was pretty simple.  The line number was the cyclist ID. The first column contains an ID for Table X, the second column contains an ID for Table Y, and so on. My PHP server has a maximum execution time of 30 seconds, so I added parameters to the script to only ingest  100 lines at a time and ran the script multiple times. Since it’s a private PHP server that doesn’t have consumer traffic, I should have just increased the timeout, let the script run, then changed it back.

While building the spreadsheet, I had been tagging photographs in Lightroom with cyclist IDs. I exported the tagged photographs into a certain directory, then wrote another PHP script to iterate through all files in that directory, read the EXIF data, and fill in the images_show_cyclists table.

New frontend

This is my process for identifying cyclists going forward.  I look at an image in Lightroom and find a new cyclist who was not in the previous image.  I may scrub back and forth in the timeline to get a better view.  I fill in the search/create page to see if I have already seen a similar cyclist.

New “clothing” dropdowns are created as existing ones are filled in, so I can specify any number of clothing items. The “description” field checks for each word in order, so “blue yellow” matches both “blue & yellow stripes on arms” and “blue torso, red arms, black legs, goofy hat, yellow face”

Clicking “Find matching cyclists” will either show a list of cyclists with the features I’ve selected, or unlock the “CREATE” button if there are no matching cyclists.  Each matching cyclist is a link that takes me to a page that lists its features, what images it appears in, and previews one of those images.

Having a picture of the cyclist on the “view cyclist” page makes it much easier to confirm if I’ve actually found the cyclist I’m looking for, since I can just look between the two images.

The “EDIT this cyclist” page is almost identical to the search/create page, but instead of starting blank, it starts with data filled in from the DB.

Cyclists make multiple laps and groups tend to stick together, so if I see one cyclist back for a second lap, I can look at photographs from her first lap and identify some of the cyclists around her as well.

Preliminary data

I haven’t examined all the photographs yet, but here are some things I’ve discovered so far.

In 2012 photos taken over 50 minutes, I identified 1475 Solstice Cyclists.

Here’s a graph of how many passed over time.  Click to expand. 1 pixel vertically = 1 cyclist. 1 pixel horizontally = 1 second.  Red represents cyclists on their first lap. Green is the second lap. Blue is the third.  There are gaps when my view was blocked, the street was empty, I had to switch memory cards, and when traffic stopped & I paused the automatic camera.

The male/female split is 49/51, even closer than Dragon Con’s demographics, and very different from the split seen in most photographers’ galleries, in which images of women dominate.  Hmmmmmm. How curious.  HMMMMMM.

1300 people rode bicycles, which is to be expected from a group called the Solstice Cyclists, but I also saw:

• 39 people on foot
• 23 on inline skates
• 6 on roller skates
• 24 on scooters
• 5 unicycles
• 10 people on 5 tandem bikes
• 7 skateboards
• 2 pedicabs, with 2 drivers & 4 passengers

I also identified some groups & popular “costumes”

• 11 giraffes
• 45 mermaids
• 8 Care Bears
• 39 people wearing actual, normal clothes
• 27 Wonder Women

I still have around 700 images to look through, so these numbers will change a bit, but as you can see from the graph, most of the cyclists in these later images are back for another lap, and there aren’t many new cyclists.

Once all that is done, I can start (START!) on the actual meat of this project: creating a grammar for bodypaint based on these thousands of examples & generating new paint patterns.

The Solstice Cyclists, an intentionally-disorganized group of mostly-naked, mostly-painted cyclists who precede and overwhelm the Fremont Solstice Parade each year are one of my favorite groups to photograph.  They are colorful. creative, joyful, and high-energy.

Last year I decided I needed a photo of every single Solstice Cyclist.  (Does this seem familiar?)  I had two reasons:

1. Statistics. Photographers’ galleries contain mostly women. Is this disparity caused by population imbalance or by selection bias? Solstice Cyclists are famous for being naked cyclists, but some people wear some clothing. How common is that?  What protective device is more common: bike helmet or sunglasses?
2. Source data for grammar. I want to expand my bodypaint generator to use graphics, and I want the generator’s output to mirror actual paintjobs. Once I identify all the different cyclists, I can study their paintjobs, break them down into parts, and put those parts back together in novel but believable ways.

I got a tripod and a timer so one camera could automatically photograph everyone who passed while I did my normal photography beside it. I had to juggle a surprising number of factors to place that camera properly.

• To avoid being blocked by spectators, the tripod needed to be either right next to the street, or high enough to shoot over their heads. I saw a few balconies, stout tree branches, even a bridge, that could get the needed height, but that brought new problems. Most paintjobs photograph best from the front, and bicycle riders tent to lean forward, so a camera that is too high has a bad angle. Also, accessing those high places is non-trivial, so I opted for a front-row seat.
• Aiming down the street at approaching cyclists is my usual MO, but an automated camera will have trouble with that.  Since the camera is looking down the street, cyclists in the same image can be 10 feet or 100 yards away. How does the camera know which one to focus on, and which ones to leave blurry?  Cyclists in front will obstruct the camera’s view of cyclists behind them.
• The route turns a few times. Maybe setting up at a corner will alleviate these issues. Setting up just after a corner sets a maximum distance at which cyclists will appear. Any further and they’d be in the crowd. There’s still the problem of cyclists approaching the camera and filling the frame, blocking other cyclists.
• What about aiming across the street?  Cyclists will stay about the same distance from the camera as they cross the frame, and they are only 3 or 4 abreast, as opposed to unlimited ranks front-to-back, so obstruction is less of an issue.  Since I’m as far forward as possible (so spectators don’t stand in front of me) cyclists on the near side of the street will be very close. My lens might not be wide enough to capture their whole bodies, and they will cross the frame very quickly, maybe in between ticks of the automatic timer.
• Thus, I decided to shoot across the street at the cyclists on the far side of the road. The frame is wide enough at that range that I’ll get several photos as each cyclist passes. Three-quarter to side view is not ideal, but still pretty good.  I had to accept cyclists on the near side sometimes blocking the shot, but it was the best I could do.
• Oh, also! Position along the parade route matters as well.  The Cyclists circle back so they stay close to the parade (human-powered floats are much slower than bicycles). Near the end of the parade route, there are fewer spectators and no returning cyclists to block my view, but I only get one chance to see each cyclist, and some cyclists leave the route before then (mechanical failures, etc.) Closer the start of the route I get multiple chances to photograph each cyclist, but more obstructions.

The day before the parade I scouted the parade route, looking for places to set up.

I chose the spot on the right, which is near the “center of the universe” sign on the east side of Fremont Ave. The tree gave some protection to the tripod. It’s a lot easier to accidentally trip over a tripod than it is to walk into a tree.

During the parade I kept looking over at the “shots remaining” counter on the tripod-mounted camera like the marines watching the sentry guns in Aliens.  “That number is going down.  It, it keeps going down.  Are we going to run out before they stop coming?”  The automatic filled a 32GB memory card and I had to swap for another in the middle of the parade.  Whenever a traffic jam stopped the stream of cyclists passing me, I’d pause the automatic camera to save disk space.

In all the automatic camera captured 2644 images.  That’s equivalent to an entire day of Atlanta Fashion Police, except it took only 63 minutes, not 16 hours.  I took an additional 1400 photos with the camera I was holding.

I considered using computer vision to help me identify cyclists, but even nudity-detecting algorithms were bamboozled by the cyclists’ coloration. So I couldn’t even get “Yes, there is a person in this photo”, much less, “There are 6 people in this photo, and the guy with the red stripes and sunglasses has appeared in 3 other photos.” Time to use my eyes, the best pattern-recognizers I know! I thought I could store all the information in a CSV file. I’m only recording a few pieces of data for each cyclist, do I really have to make an SQL database with webforms to search and update it?

1064 rows later, I realized that, yes, I did need that DB.  Since cyclists could make several laps, and I was gathering data from both cameras, I needed to check for duplicate cyclists often.  Ctrl-F in a spreadsheet wasn’t cutting it.

Next time: building that database, and a few insights from the data.