Well, this isn't good...
#1
What happened?
Thread Starter
Join Date: Jun 2007
Location: Around here somewhere
Posts: 7,927
Bikes: 3 Rollfasts, 3 Schwinns, a Shelby and a Higgins Flightliner in a pear tree!
Mentioned: 57 Post(s)
Tagged: 1 Thread(s)
Quoted: 1835 Post(s)
Liked 292 Times
in
255 Posts
Well, this isn't good...
https://www.vox.com/future-perfect/2...icle-dark-skin
So now autonomous cars are also 'racially biased'? Not in a conventional sense, but they have trouble seeing dark skinned pedestrians.
But they can see dark cars?
A bit crazy.
So now autonomous cars are also 'racially biased'? Not in a conventional sense, but they have trouble seeing dark skinned pedestrians.
But they can see dark cars?
A bit crazy.
__________________
I don't know nothing, and I memorized it in school and got this here paper I'm proud of to show it.
#3
Senior Member
Join Date: Oct 2015
Posts: 15,491
Bikes: 2015 Workswell 066, 2017 Workswell 093, 2014 Dawes Sheila, 1983 Cannondale 500, 1984 Raleigh Olympian, 2007 Cannondale Rize 4, 2017 Fuji Sportif 1 LE
Mentioned: 144 Post(s)
Tagged: 0 Thread(s)
Quoted: 7652 Post(s)
Liked 3,478 Times
in
1,836 Posts
Well .... one solution would be to make pedestrians wear clothes.
#4
Senior Member
Join Date: Oct 2015
Posts: 15,491
Bikes: 2015 Workswell 066, 2017 Workswell 093, 2014 Dawes Sheila, 1983 Cannondale 500, 1984 Raleigh Olympian, 2007 Cannondale Rize 4, 2017 Fuji Sportif 1 LE
Mentioned: 144 Post(s)
Tagged: 0 Thread(s)
Quoted: 7652 Post(s)
Liked 3,478 Times
in
1,836 Posts
"The report, “Predictive Inequity in Object Detection,” should be taken with a grain of salt. It hasn’t yet been peer-reviewed. It didn’t test any object-detection models actually being used by self-driving cars, nor did it leverage any training datasets actually being used by autonomous vehicle manufacturers."
Basically they made up their own detection device. it didn't work, so they decided no one else's did either.
Basically they made up their own detection device. it didn't work, so they decided no one else's did either.
#5
Senior Member
Join Date: Sep 2013
Location: Massachusetts
Posts: 4,530
Mentioned: 34 Post(s)
Tagged: 0 Thread(s)
Quoted: 2112 Post(s)
Liked 663 Times
in
443 Posts
"The report, “Predictive Inequity in Object Detection,” should be taken with a grain of salt. It hasn’t yet been peer-reviewed. It didn’t test any object-detection models actually being used by self-driving cars, nor did it leverage any training datasets actually being used by autonomous vehicle manufacturers."
Basically they made up their own detection device. it didn't work, so they decided no one else's did either.
Basically they made up their own detection device. it didn't work, so they decided no one else's did either.
"Nevertheless, recent, state-of-the art systems designed by many major tech conglomerates have continued to face scrutiny for the behavior of their facial recognition systems."
It should come as no surprise that there might be a disparity in the domain of recognizing pedestrians as well.
Why they did not test with "object-detection models actually being used by self-driving cars" or "training datasets actually being used by autonomous vehicle manufacturers?”
- Proprietary object-detection models
- Proprietary training/validation datasets
So they could have just FULL STOP. Well, that's that then. Trust, don't validate.
Or, they could :
- Annotate an open training/validation dataset
- "The Berkeley Driving Dataset (Yu et al., 2018) is one of the most comprehensive driving datasets to date."
- Train and validate open detection models
- "In this paper, we study the performance of several models used in state-of-the art (He et al., 2017) object detection, and show uniformly poorer performance of these systems when detecting pedestrians with Fitzpatrick skin types between 4 and 6."
- Time of day and occlusion did *NOT* explain the disparity.
- "Nonetheless, this data suggests an important conclusion that there is not evidence that time of day is to blame for the predictive inequity observed overall."
- "Therefore, we conclude that the source of discrepancy... is not due to co-occurrence with occluded people."
- But the disparity could be partially mitigated by re-weighting loss functions.
- "This result suggests that careful reweighting may improve performance on DS pedestrians without sacrificing performance on LS pedestrians, or even improve performance on LS pedestrians as a byproduct."
Or not. After all, people of color can wear long sleeve shirts, white gloves, and hi viz socks, and all will be well, maybe?
-mr. bill
Last edited by mr_bill; 03-07-19 at 07:09 PM.
#6
On Holiday
Join Date: Apr 2017
Posts: 1,014
Bikes: A bunch of old steel bikes
Mentioned: 14 Post(s)
Tagged: 0 Thread(s)
Quoted: 394 Post(s)
Likes: 0
Liked 14 Times
in
12 Posts
Another "newsy" announcement about a hobbled non-peer reviewed study based on iffy data. If the intent was to stir things up - mission accomplished.
#7
Senior Member
Join Date: Jan 2009
Location: Colorado Springs, CO
Posts: 5,974
Bikes: 2015 Charge Plug, 2007 Dahon Boardwalk, 1997 Nishiki Blazer, 1984 Nishiki International, 2006 Felt F65, 1989 Dahon Getaway V
Mentioned: 54 Post(s)
Tagged: 0 Thread(s)
Quoted: 1364 Post(s)
Liked 1,677 Times
in
827 Posts
Reminds me off this:
#8
• —
Join Date: Jun 2015
Location: Land of Pleasant Living
Posts: 12,241
Bikes: Shmikes
Mentioned: 59 Post(s)
Tagged: 0 Thread(s)
Quoted: 10171 Post(s)
Liked 5,867 Times
in
3,157 Posts
The designers have to get that ~2:1 passing ratio right. Dangerous to put vehicles that behave atypically out there.
#9
Senior Member
Join Date: Jan 2010
Posts: 39,243
Mentioned: 211 Post(s)
Tagged: 1 Thread(s)
Quoted: 18417 Post(s)
Liked 15,557 Times
in
7,332 Posts
#10
Senior Member
Join Date: Oct 2015
Posts: 15,491
Bikes: 2015 Workswell 066, 2017 Workswell 093, 2014 Dawes Sheila, 1983 Cannondale 500, 1984 Raleigh Olympian, 2007 Cannondale Rize 4, 2017 Fuji Sportif 1 LE
Mentioned: 144 Post(s)
Tagged: 0 Thread(s)
Quoted: 7652 Post(s)
Liked 3,478 Times
in
1,836 Posts
"Tesla says the autopilot function is intended to increase safety on the roads." and also makes it easier to plow into parked fire trucks.
#11
Senior Member
Join Date: Jul 2005
Posts: 5,992
Mentioned: 26 Post(s)
Tagged: 0 Thread(s)
Quoted: 2495 Post(s)
Liked 738 Times
in
522 Posts
Autopilots in aircraft at cruising altitude in monitored airspace are one thing, autopilots in autonomous vehicles operating on public roads ... ... oh the hubris! What were they smoking? Countless people are paralyzed from the neck down because they took their eyes off the prize for less time than it takes to say "pardonez moi, do you have any WHAP ... things happen fast out among the heathen and as good as cybernetic neural nets are in the lab they have proven to have blindspots (swidt?) when turned loose in full autonomous mode. Surely, for many years to come, the successful implementation of AI guidance of ground based motor vehicles will be as 'driver assist' systems using the effective sensor suites of the full autonomous systems to augment human intelligence? There are presently drowsy driver detection systems, blindspot monitoring systems, and close vehicle collision warning systems in place in regular production automobiles and these are being expanded upon with out of lane warnings and other semi-automatic guidance systems. Full on AI is NOT intended to keep cyclists safe! Full on AI is intended to replace fleet operators (including taxi and limousine operators) so that owners of trucking and rideshare concerns can reduce their labor costs just that of mechanics and administrative personnel. AV is an answer to a question no rational person asked. We should probably fight their adoption.
#12
Senior Member
Join Date: Oct 2015
Posts: 15,491
Bikes: 2015 Workswell 066, 2017 Workswell 093, 2014 Dawes Sheila, 1983 Cannondale 500, 1984 Raleigh Olympian, 2007 Cannondale Rize 4, 2017 Fuji Sportif 1 LE
Mentioned: 144 Post(s)
Tagged: 0 Thread(s)
Quoted: 7652 Post(s)
Liked 3,478 Times
in
1,836 Posts
You know why I hate "studies" like this one? because this is what they are Really all about: Self-Driving Cars Are More Likely To Run You Over If You Are Black, Study Suggests (https://www.iflscience.com/technolog...PHSONtemHGaTWI)
I see this all day long, everyday. Scientists looking to get publicity to help get future funding release some lame, half-baked "study" with a questionable conclusion and then some publicist rights a click-bait headline which totally obscure the fact that the study didn't actually test any of the systems in use, not any explanation of Why the data seemed to show what it did. Some headline writing editor whose Job is the Create Click-Bait headlines, not to tell the truth, then warps the message even further, suggesting a programmed-in bias against people of color.
Now there will be millions of people saying "AVs are designed to kill blacks and Hispanics."
My question is still----since almost everyone is Clothed when out in public, how is this working or not working exactly (of course, no one knows whether the current systems even do this, but .... ) Does the color of the clothes makes a difference? Do people wearing gloves and hats get noticed more or less? What exactly leads to this outcome? No one knows. But .... the publicist for this group made sure the headline was sensational, and now that sensationalized headline is used to get clicks on websites---while glossing over the fact that no one knows if the systems being used by the big AV manufacturers share the same problems.
I see this all day long, everyday. Scientists looking to get publicity to help get future funding release some lame, half-baked "study" with a questionable conclusion and then some publicist rights a click-bait headline which totally obscure the fact that the study didn't actually test any of the systems in use, not any explanation of Why the data seemed to show what it did. Some headline writing editor whose Job is the Create Click-Bait headlines, not to tell the truth, then warps the message even further, suggesting a programmed-in bias against people of color.
Now there will be millions of people saying "AVs are designed to kill blacks and Hispanics."
My question is still----since almost everyone is Clothed when out in public, how is this working or not working exactly (of course, no one knows whether the current systems even do this, but .... ) Does the color of the clothes makes a difference? Do people wearing gloves and hats get noticed more or less? What exactly leads to this outcome? No one knows. But .... the publicist for this group made sure the headline was sensational, and now that sensationalized headline is used to get clicks on websites---while glossing over the fact that no one knows if the systems being used by the big AV manufacturers share the same problems.
#13
Senior Member
Join Date: Sep 2013
Location: Massachusetts
Posts: 4,530
Mentioned: 34 Post(s)
Tagged: 0 Thread(s)
Quoted: 2112 Post(s)
Liked 663 Times
in
443 Posts
Can you still see your skin? How is that possible? You are dressed to go out in public!
It’s trivial btw to annotate the dataset to include color of clothes, coverage, and so on.
See BDD100K.
To run your own simulations, use the source, Luke.
Looking forward to hearing what you learn.
-mr. bill
Last edited by mr_bill; 03-08-19 at 06:34 AM.
#14
Senior Member
Join Date: Jan 2010
Posts: 39,243
Mentioned: 211 Post(s)
Tagged: 1 Thread(s)
Quoted: 18417 Post(s)
Liked 15,557 Times
in
7,332 Posts
#15
Senior Member
Join Date: Aug 2017
Location: Mid Atlantic / USA
Posts: 2,115
Bikes: 2017 Specialized Crosstrail / 2013 Trek Crossrip Elite
Mentioned: 43 Post(s)
Tagged: 0 Thread(s)
Quoted: 1002 Post(s)
Liked 237 Times
in
155 Posts
This is EXACTLY what I thought of too when I saw the OP!!! HAHAHAHA
Better Off Ted was a short lives single season comedy. It was okay...but that one episode was one of the funniest half hours of comedy in television history.
Better Off Ted was a short lives single season comedy. It was okay...but that one episode was one of the funniest half hours of comedy in television history.
Reminds me off this:
https://www.youtube.com/watch?v=lMy5YpJysy4
https://www.youtube.com/watch?v=lMy5YpJysy4
#16
Senior Member
Join Date: Sep 2013
Location: Massachusetts
Posts: 4,530
Mentioned: 34 Post(s)
Tagged: 0 Thread(s)
Quoted: 2112 Post(s)
Liked 663 Times
in
443 Posts
BTW, how stuff like this goes viral in the popular press, see what happened to BDD100K.
Still want to blindly attribute malice to geeks?
Oh, besides that backstory, note "Our dataset is crowd-sourced, and covers a very large area and diverse visual phenomena (indeed significantly more diverse than previous efforts, in our view), but it is very clearly limited to monocular RGB image data and associated mobile device metadata."
-mr. bill
Still want to blindly attribute malice to geeks?
Oh, besides that backstory, note "Our dataset is crowd-sourced, and covers a very large area and diverse visual phenomena (indeed significantly more diverse than previous efforts, in our view), but it is very clearly limited to monocular RGB image data and associated mobile device metadata."
-mr. bill
Last edited by mr_bill; 03-08-19 at 08:16 AM.
#17
Senior Member
Join Date: Sep 2008
Location: NE Tennessee
Posts: 917
Bikes: Giant TCR/Surly Karate Monkey/Foundry FireTower/Curtlo Tandem
Mentioned: 4 Post(s)
Tagged: 0 Thread(s)
Quoted: 169 Post(s)
Liked 84 Times
in
62 Posts
Why does this make me thing of National Lampoon's vacation where Clark fell asleep at the wheel.
#18
Senior Member
Join Date: Apr 2005
Location: Ann Arbor, MI
Posts: 16,878
Bikes: 1980 Masi, 1984 Mondonico, 1984 Trek 610, 1980 Woodrup Giro, 2005 Mondonico Futura Leggera ELOS, 1967 PX10E, 1971 Peugeot UO-8
Mentioned: 49 Post(s)
Tagged: 0 Thread(s)
Quoted: 1858 Post(s)
Liked 664 Times
in
506 Posts
Just to be clear, this study did not say that vehicles will show this biased behavior. It shows (IMO) that biased behavior is possible.
The question for vehicles is, will cars accurately detect all objects actually located in the field of traffic, classify all the pedestrians as such without errors, and determine the relevant aspects of the absolute and relative motion of the pedestrians to determine if they are placed at hazard by the motion of the host vehicle. I don't think this can be assessed with accuracy when the test did not involve any actual prototype or production vehicle sensors, did not involve any vehicle-based vision-processing and object detection/classification algorithms, and did not use any actual automotive object processing.
The car does not need to identify whom has been detected (Congress person versus convict). It does need to determine if it is a person, if it is in-path or when it will be in-path, and how the risk of collision changes with relative vehicle-to-object motion. That's not the problem these researchers set for themselves.
Other vision systems in cars look at the driver (for SAE automation levels 4 and below) to determine alertness, direction of gaze, and other characteristics that could assess safety of driver actions. While these look at the face, they also do not need to recognize who the driver is nor the race, gender or size. Race, gender, size, clothing, hairstyle, cleanliness, cosmetics, health, morphology variations natural or not, eyewear and possibly other points affect these determinations, but the basic requirements are that the head pose, eyelid status, possible gaze direction, and eye motion be detected. Not identity.
Sometimes the real problem is easier than what the researchers have available to them.
But, it is necessary to impose design and compliance/verification/validation testing requirements for such systems to direct the designer (including his/her company) to ensure no loss of relevant sensor performance exists, and to verify that the performance is effectively equal across the population. In the United States the only regulatory agency that is able to constrain automotive design is NHTSA, and their Congressional mandate only permits them to act reactively. The US structure of liability law values that safety-related design must follow state-of-the-art practices to ensure safety, or due diligence might not have been followed. This usually motivates companies within the US auto industry to "do the right thing," but ... it's not a regulation. Europe mandates following state-of-the-art practices and verifying that such has in fact been done, achieving firm regulation to their standards.
Ensuring equality across the population of design teams does not necessarily solve the problem (how do you make sure that all design teams have a member who has had significant facial injuries, for just one example requirement?), because all industrial technical professionals are subject to pressure from the corporate structure. An engineer from India, one from Africa, one from USA, and one from Asia are all subjected to cost and timing pressure when performing product design. And all are capable of eliminating biases from their thinking in their work, if they actually have any biases. And being human, imperfections can creep in despite best efforts of all involved.
This paper shows a potential for biased behavior and that training methodology may help to remove it, but it is not directly applicable for vehicles as it is.
#19
Senior Member
Join Date: Oct 2013
Location: Oahu, HI
Posts: 1,396
Bikes: 89 Paramount OS 84 Fuji Touring Series III New! 2013 Focus Izalco Ergoride
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 285 Post(s)
Likes: 0
Liked 74 Times
in
54 Posts
I've seen a few critiques lately of training sets used for AI resulting in AI "learning" the wrong things.
scott s.
.
scott s.
.
#20
Senior Member
Join Date: Feb 2019
Location: Indiana
Posts: 592
Bikes: 1984 Fuji Club, Suntour ARX; 2013 Lynskey Peloton, mostly 105 with Ultegra rear derailleur, Enve 2.0 fork; 2020 Masi Giramondo 700c, full Deore with TRP dual piston mech disk brakes
Mentioned: 5 Post(s)
Tagged: 0 Thread(s)
Quoted: 324 Post(s)
Liked 81 Times
in
71 Posts
https://www.vox.com/future-perfect/2...icle-dark-skin
So now autonomous cars are also 'racially biased'? Not in a conventional sense, but they have trouble seeing dark skinned pedestrians.
But they can see dark cars?
A bit crazy.
So now autonomous cars are also 'racially biased'? Not in a conventional sense, but they have trouble seeing dark skinned pedestrians.
But they can see dark cars?
A bit crazy.
https://www.popularmechanics.com/car...rs-less-black/
Last edited by greatscott; 03-09-19 at 07:33 AM.
#21
Senior Member
Join Date: Oct 2015
Posts: 15,491
Bikes: 2015 Workswell 066, 2017 Workswell 093, 2014 Dawes Sheila, 1983 Cannondale 500, 1984 Raleigh Olympian, 2007 Cannondale Rize 4, 2017 Fuji Sportif 1 LE
Mentioned: 144 Post(s)
Tagged: 0 Thread(s)
Quoted: 7652 Post(s)
Liked 3,478 Times
in
1,836 Posts
What they're not telling you is that these systems can't see black cars either, ....
https://www.popularmechanics.com/car...rs-less-black/
https://www.popularmechanics.com/car...rs-less-black/
Well, hey, if AVs reduce accidents by 5/6 I will be happy. I don't own any black jerseys or blacked-out bikes.
#22
Senior Member
Join Date: Feb 2019
Location: Indiana
Posts: 592
Bikes: 1984 Fuji Club, Suntour ARX; 2013 Lynskey Peloton, mostly 105 with Ultegra rear derailleur, Enve 2.0 fork; 2020 Masi Giramondo 700c, full Deore with TRP dual piston mech disk brakes
Mentioned: 5 Post(s)
Tagged: 0 Thread(s)
Quoted: 324 Post(s)
Liked 81 Times
in
71 Posts
Of course with todays idiots driving and texting a car with a stopping system is a good idea, but why should WE pay to have that stopping system or autonomous systems in our cars when all they really have to do, and it's a heck of a lot cheaper, is to make all cell phones shut off automatically at speeds greater than 15 mph, and actually the technology to do that is already in the phone! And the only number you could enter into the phone and have it work at speeds over 15 is 911. We pay far too much for cars now, and these damn autonomous systems are making them even more expensive, all so we can yak and text on a phone...it's not worth it, we all got a long just fine driving our cars without the need of wanting to be on the phone the whole time driving, I'm sure we could get along fine without them again.
By the way, this autonomous system was also having troubles "seeing" cyclists...I can assume the rich corporate types that created this system hate cyclists as well.
Last edited by greatscott; 03-09-19 at 07:37 AM.
#23
Senior Member
Join Date: May 2009
Location: Tampa/St. Pete, Florida
Posts: 9,352
Bikes: Specialized Hardrock Mountain (Stolen); Giant Seek 2 (Stolen); Diamondback Ascent mid 1980 - 1997
Mentioned: 4 Post(s)
Tagged: 0 Thread(s)
Quoted: 62 Post(s)
Likes: 0
Liked 3 Times
in
3 Posts
https://www.vox.com/future-perfect/2...icle-dark-skin
So now autonomous cars are also 'racially biased'? Not in a conventional sense, but they have trouble seeing dark skinned pedestrians.
But they can see dark cars?
A bit crazy.
So now autonomous cars are also 'racially biased'? Not in a conventional sense, but they have trouble seeing dark skinned pedestrians.
But they can see dark cars?
A bit crazy.
About the only thing that I agree with him with regards to fully autonomous self-driving cars, is that we as cyclists and pedestrians shouldn't have to wear any sort of transponder so that they'll know where we are.
He is also very surprised that people are still talking about the failures from companies like Tesla and Uber instead of all of the successes.
#24
Senior Member
Join Date: Feb 2019
Location: Indiana
Posts: 592
Bikes: 1984 Fuji Club, Suntour ARX; 2013 Lynskey Peloton, mostly 105 with Ultegra rear derailleur, Enve 2.0 fork; 2020 Masi Giramondo 700c, full Deore with TRP dual piston mech disk brakes
Mentioned: 5 Post(s)
Tagged: 0 Thread(s)
Quoted: 324 Post(s)
Liked 81 Times
in
71 Posts
I know someone who swears up and down that fully autonomous self-driving cars are not only the "wave of the future," BUT that they'll be here LONG before the actual experts have predicted that they'll be here. Not only that but he thinks that they'll "replace" public/mass transit, and that they'll be a boon to both cyclist, but pedestrian safety.
About the only thing that I agree with him with regards to fully autonomous self-driving cars, is that we as cyclists and pedestrians shouldn't have to wear any sort of transponder so that they'll know where we are.
He is also very surprised that people are still talking about the failures from companies like Tesla and Uber instead of all of the successes.
About the only thing that I agree with him with regards to fully autonomous self-driving cars, is that we as cyclists and pedestrians shouldn't have to wear any sort of transponder so that they'll know where we are.
He is also very surprised that people are still talking about the failures from companies like Tesla and Uber instead of all of the successes.
#25
Been Around Awhile
Join Date: Oct 2004
Location: Burlington Iowa
Posts: 29,973
Bikes: Vaterland and Ragazzi
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 12 Post(s)
Liked 1,536 Times
in
1,045 Posts
https://www.cnbc.com/2018/05/24/uber...cials-say.html