lunitiks
Cool James & Black Teacher
And good lectures!Amnon Shashua has good lecturers on this.
You can install our site as a web app on your iOS device by utilizing the Add to Home Screen feature in Safari. Please see this thread for more details on this.
Note: This feature may not be available in some browsers.
And good lectures!Amnon Shashua has good lecturers on this.
I guess I'll post here too.
The latest snapshotting logic shows us that they actually have some quite advanced labels.
A snapshot trigger for my car requesting "label: construction" triggered for my car today and created this image:
There's a bunch of other labels requested like "Confusing lanes", "slope up/down", "barrier" and so on.
I guess I have an urgent need to drive around in strange places to see what else I can trigger
Mind blowing stuff!
So this must be training in action, right? Your car obviously ‘knows’ (or rather, suspects) when it sees a roadblock, and then uploads the img to mothership and they do the validation thing?
The snapshot thing is a very recent development, I literally got the trigger for this snapshot last night, so it's not something they request from all cars at all times.We've observed for some time now that the steering wheel prompt behavior on AP1 and AP2 changes when there are construction barrels in the camera's view (prompts much more often.) Interesting to see it's sending snapshots as well.
The snapshot thing is a very recent development, I literally got the trigger for this snapshot last night, so it's not something they request from all cars at all times.
NN obviously is trained to recognize certain things, so one of those things is "construction zone" I guess. So then they request snapshots of the matching labels.How does it know it's a construction site? Is the NN identifying it? That's bananas.
{"query":{"$and":[{"$eq":["@LegacyDebug.gear",4]},
{"$gt":["@TelemetryOutput.distance_travelled_m",2000]},
{"$gt":["@LegacyDebug.veh_speed_mps",5]},{"$lt":["@LegacyDebug.veh_speed_mps",80]},
{"$gt":[{"$sum":[{"$labelboost":"$BIG_4K_of_data","version":"0.1","camera":"main"},
-0.012629894464]},0.5]},{"$gt":["@VisionImageEmbeddings.main.timestamp_ns",0]}]}
NN obviously is trained to recognize certain things, so one of those things is "construction zone" I guess. So then they request snapshots of the matching labels.
The snapshot request here looks like this:
How does it know it's a construction site? Is the NN identifying it? That's bananas.
Follow up: which cameras is it using to identify these label areas, all or just the forward / repeaters?
I guess I'll post here too.
The latest snapshotting logic shows us that they actually have some quite advanced labels.
A snapshot trigger for my car requesting "label: construction" triggered for my car today and created this image:
There's a bunch of other labels requested like "Confusing lanes", "slope up/down", "barrier" and so on.
I guess I have an urgent need to drive around in strange places to see what else I can trigger
Seems like they're requesting pictures to ensure the NN is correct identifying these things.
Yep, agree. Probably not so much training than validation.
It would flag false positives (or rather whoever does later review), but not false negatives.Probably a bit of both.... the problem is if their identify is off it doesn’t take a snapshot of an edge case right?
they only send the main one for the label requests.
The curve fitting now sends 4 cameras: main, fisheye and the B pillars (both)
That's really fascinating. The B pillar cams sure seem to be the best ones for determining if your car is currently centered in your lane or not. The main camera looks pretty far ahead, so this could potentially be for validation that the new control algorithm is performing acceptably.
Damn, looks like only on occasion did these cameras show lane lines. I’m inclined to agree with you, do not seem like they’re the best for validation. Not sure what they were hoping with regards to curves, compared to the fisheye.I didn't think they were wide enough angle to show you the lane lines right next to you. I would have thought the "repeater" cameras on the front fenders facing backwards are a better choice for seeing lane centering.