YOLOv5: State-of-the-art object detection at 140 FPS • Hacker News Board

Re: YOLOv5: State-of-the-art object detection at 140 FPS

#21

Post by rocauc » Wed, Jun 10, 2020, 5:09 PM UTC

sillysaurusx wrote: post #18

We made a site that lets you collaboratively tag a bunch of images, called tagpls.com. For example, users decided to re-tag imagenet for fun: https://twitter.com/theshawwn/status/1262535747975868418 And the tags ended up being hilarious: https://pbs.twimg.com/media/EYXRzDAUwAMjXIG?format=jpg&name=... (I'm particularly fond of https://i.imgur.com/ZMz2yUc.png ) The data is freely available via API: https://www.tagpls.c…

I remember following this as it came out (and learning windshield wipers should be called "swipey bois")

Surprised and happy to hear you're seeing high labeling quality.

We'll re-host with credit on https://public.roboflow.ai What license is this?

Re: YOLOv5: State-of-the-art object detection at 140 FPS

#22

Post by qchris » Wed, Jun 10, 2020, 5:21 PM UTC

If anyone's interested in the direct GitHub link to the repository: https://github.com/ultralytics/yolov5

Re: YOLOv5: State-of-the-art object detection at 140 FPS

#23

Post by jcims » Wed, Jun 10, 2020, 5:31 PM UTC

ahelwer wrote: post #14

osipov wrote: post #10

Just recently IBM announced with a loud PR move that the company is getting out of the face recognition business. Guess what? Wall Street doesn't want to keep subsidizing IBM's subpar face recognition technology when open source and Google solutions are pushing the state of the art.

Not something to brag about. Facial recognition has very few applications outside of total surveillance. We should not respect those who lend it their time and effort.

>Facial recognition has very few applications outside of total surveillance.

That's not really for you to decide, is it? You're absolutely free to have that opinion of course.

>We should not respect those who lend it their time and effort.

Also your choice of course. Facial recognition is essentially a light integration of powerful underlying technologies. Should 'we' ostracize those working on machine learning, computer vision, network and distributed computing, etc?

Re: YOLOv5: State-of-the-art object detection at 140 FPS

#24

Post by sillysaurusx » Wed, Jun 10, 2020, 5:32 PM UTC

rocauc wrote: post #21

sillysaurusx wrote: post #18

We made a site that lets you collaboratively tag a bunch of images, called tagpls.com. For example, users decided to re-tag imagenet for fun: https://twitter.com/theshawwn/status/1262535747975868418 And the tags ended up being hilarious: https://pbs.twimg.com/media/EYXRzDAUwAMjXIG?format=jpg&name=... (I'm particularly fond of https://i.imgur.com/ZMz2yUc.png ) The data is freely available via API: https://www.tagpls.c…

I remember following this as it came out (and learning windshield wipers should be called "swipey bois") Surprised and happy to hear you're seeing high labeling quality. We'll re-host with credit on https://public.roboflow.ai What license is this?

Thanks! We've decided to license the data as CC-0. We'll add that to the footer.

We don't host any images directly – we merely serve a list of URLs (e.g. https://battle.shawwn.com/tfdne.txt). But any data served via the API endpoints is CC-0.

Re: YOLOv5: State-of-the-art object detection at 140 FPS

#25

Post by anewdirection » Wed, Jun 10, 2020, 5:32 PM UTC

monocasa wrote: post #19

ahelwer wrote: post #14

Earlier quoted context omitted.

Not something to brag about. Facial recognition has very few applications outside of total surveillance. We should not respect those who lend it their time and effort.

I thought the real focus on the bad actors at this point was on gait detection. Works in civil unrest situations where everyone covers their face. Not the the difference matters that much.

Its not exclusive. Bad actors are working on whatever they are paid to build, by other bad actors with less technical acumen and more money.

Edit: I should add, that most of the actual progress is being made by smart people who think its an interesting problem and are unaware or uncaring of the clear outcome of such tech.

Re: YOLOv5: State-of-the-art object detection at 140 FPS

#26

Post by sillysaurusx » Wed, Jun 10, 2020, 5:35 PM UTC

sillysaurusx wrote: post #18

We made a site that lets you collaboratively tag a bunch of images, called tagpls.com. For example, users decided to re-tag imagenet for fun: https://twitter.com/theshawwn/status/1262535747975868418 And the tags ended up being hilarious: https://pbs.twimg.com/media/EYXRzDAUwAMjXIG?format=jpg&name=... (I'm particularly fond of https://i.imgur.com/ZMz2yUc.png ) The data is freely available via API: https://www.tagpls.c…

Well, that didn't take long – our API endpoint keeled over. Luckily, you can fetch all the data directly from firebase:

  # fetch raw tag data
  $ curl -fsSL https://experiments-573d7.firebaseio.com/results/.json > tags.json
  $ du -hs tags.json
  14M tags.json

  # fetch tag metadata (colors, remapping label names, possibly other stuff in the future)
  $ curl -fsSL https://experiments-573d7.firebaseio.com/user_meta/.json > tags_meta.json
  $ du -hs tags_meta.json
  376K tags_meta.json
  $ jq tags_meta.json

Note that's the raw unprocessed data (no yolo). To get info about all experiments, you can use this:

  curl -fsSL https://experiments-573d7.firebaseio.com/meta/.json | jq

I'm a bit worried about the bill. It's up to $50 and rising: https://imgur.com/ZgmXsWU almost entirely egress bandwidth. Be gentle with those `curl` statements. :)

(I think that's due to a poor architectural decision on my part, which is solvable, and not due to egress bandwidth via the API endpoint. But it's always fun to see a J curve in your bill... It's about $1 a day right now. https://imgur.com/4gUTLO7)

Re: YOLOv5: State-of-the-art object detection at 140 FPS

#27

Post by ely-s » Wed, Jun 10, 2020, 5:36 PM UTC

There seems to be an unfair comparison between the various network architectures. The reported speed and accuracy improvements should be taken with a bit of scepticism for two reasons.

* This is the first yolo implemented in Pytorch. Pytorch is the fastest ml framework around, so some of YOLOv5's speed improvements may be attributed to the platform it was implemented on rather than actual scientific advances. Previous yolos were implemented using darknet, and EfficientDet is implemented in TensorFlow. It would be necessary to train them all on the same platform for a fair speed comparison.

* EfficientDet was trained on the 90-class COCO challenge (1), while YOLOv5 was trained on 80 classes (2).

[1] https://github.com/ultralytics/yolov5/blob/master/data/coco....

[2] https://github.com/google/automl/blob/master/efficientdet/in...

Re: YOLOv5: State-of-the-art object detection at 140 FPS

#28

Post by hopfscotch2 » Wed, Jun 10, 2020, 5:39 PM UTC

jcims wrote: post #23

ahelwer wrote: post #14

Earlier quoted context omitted.

Not something to brag about. Facial recognition has very few applications outside of total surveillance. We should not respect those who lend it their time and effort.

>Facial recognition has very few applications outside of total surveillance. That's not really for you to decide, is it? You're absolutely free to have that opinion of course. >We should not respect those who lend it their time and effort. Also your choice of course. Facial recognition is essentially a light integration of powerful underlying technologies. Should 'we' ostracize those working on machine learning, comp…

If they want to use it to enable surveillance, yes.

Re: YOLOv5: State-of-the-art object detection at 140 FPS

#29

Post by TehCorwiz » Wed, Jun 10, 2020, 5:42 PM UTC

jcims wrote: post #23

ahelwer wrote: post #14

Earlier quoted context omitted.

Not something to brag about. Facial recognition has very few applications outside of total surveillance. We should not respect those who lend it their time and effort.

>Facial recognition has very few applications outside of total surveillance. That's not really for you to decide, is it? You're absolutely free to have that opinion of course. >We should not respect those who lend it their time and effort. Also your choice of course. Facial recognition is essentially a light integration of powerful underlying technologies. Should 'we' ostracize those working on machine learning, comp…

You didn't really address the author's point which was that there don't appear to be compelling uses of facial technology beyond mass automated surveillance.

I can't think of other uses and I'd be interested if you can come up with some.

Re: YOLOv5: State-of-the-art object detection at 140 FPS

#30

Post by lm28469 » Wed, Jun 10, 2020, 5:43 PM UTC

jcims wrote: post #23

ahelwer wrote: post #14

Earlier quoted context omitted.

Not something to brag about. Facial recognition has very few applications outside of total surveillance. We should not respect those who lend it their time and effort.

>Facial recognition has very few applications outside of total surveillance. That's not really for you to decide, is it? You're absolutely free to have that opinion of course. >We should not respect those who lend it their time and effort. Also your choice of course. Facial recognition is essentially a light integration of powerful underlying technologies. Should 'we' ostracize those working on machine learning, comp…

The question is always the same: is every technical/scientific progress desirable ? But it seems that this question isn't asked anymore, "move fast and break things" am I right ?

I'm much more worried about people using your arguments to try and shut down the discussion than people trying to open the debate, because once the mass surveillance/face recognition mass adoption pandora's box is open there won't be any way to go back.

When I see predator drones and FBI stingray planes above every major us cities during protests I already know we're not going in the "let's talk about this before reaching the point no return" direction.