Rendered at 00:33:58 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
patches11 1 hours ago [-]
Very interesting, I work in a similar space with diverse cameras and we’ve been using DepthAnything for a while, but I hadn’t seen these new models yet so thanks!
The association model seems like the special sauce, anything more you can share about that? Architecture, inputs and outputs etc. I’m always really interested in unique CV models.
nils_spatial 39 minutes ago [-]
Yeah, DepthAnything (especially DA3) is also really good. The space is moving so quickly, and I always try to keep an eye on the trending HF depth estimation models :D
We frame the association model as an instance segmentation ranking problem. Given a set of freight item segmentations, the question becomes which freight item is most likely being scanned? Our model needs to analyze the pose of the workers, consider multiple frames and map the whole scene to 3D.
dtrav 3 hours ago [-]
I fully get this and think its an excellent piece of work. Have you considered interfacing into Warehouse Management in order to provide dimensioned arbitrary pallet heights ? In other words to inform the put away process in warehousing ?
nils_spatial 2 hours ago [-]
Thanks!! For now, we are focusing on LTL, but improving the put away process in warehousing with better dimensioning data would be a great next use case. thanks for your thoughts!
bashd4 7 hours ago [-]
This is really cool. Congrats guys. Just because I'm curious, how does the market for this look? How much revenue are you saving your customers with this? Also, surely this is very applicable in many many industries, do you have expansion plans?
Nevertheless, this is awesome and I wish I'd built it :)
nils_spatial 5 hours ago [-]
thanks!! Our wedge is underbilling in LTL trucking with ~10k relevant cross-docking warehouses across the US and Europe. Carriers lose revenue when shippers understate freight dimensions. We're seeing ~$50k/site/month in recoverable revenue from fixing that alone.
And yes, from there we expand to other industries such as fulfillment and manufacturing. Long term, we will be the CV layer for any warehouse running CCTV.
7 hours ago [-]
pX0r 7 hours ago [-]
Interesting app of CV in OR.
Questions:
- what is the measurement precision?
- do you need calibration? How do you do it in production?
- what it is the root problem you are trying to solve?
- what is your hypothesis about your solution- quantitatively?
nils_spatial 6 hours ago [-]
1 + 4) If the bbox fit is accurate, we are below 1.5 inch MAE today. Improving bbox fit accuracy is where most of our effort goes. We're confident this gets to <1 inch at full coverage. The tail is bounded by data and model scale, both of which we're actively closing.
2) Not necessarily. Models like MapAnything/MoGe predict calibration params directly and GeoCalib is good for distortion coefficients. We still calibrate manually on-site, but mainly to validate these models actually hold up in real warehouses and collect our own calibration training data. We are confident the future is calibration-free.
3) carriers lose money every day because shippers understate dimensions and LTL is priced by volume. Every understated shipment is lost revenue. Thats the wedge we sre going after.
snihalani 8 hours ago [-]
Error Code: SSL_ERROR_RX_RECORD_TOO_LONG
nils_spatial 7 hours ago [-]
Oh shoot - where did it happen? Can you try again?
The association model seems like the special sauce, anything more you can share about that? Architecture, inputs and outputs etc. I’m always really interested in unique CV models.
We frame the association model as an instance segmentation ranking problem. Given a set of freight item segmentations, the question becomes which freight item is most likely being scanned? Our model needs to analyze the pose of the workers, consider multiple frames and map the whole scene to 3D.
Nevertheless, this is awesome and I wish I'd built it :)
And yes, from there we expand to other industries such as fulfillment and manufacturing. Long term, we will be the CV layer for any warehouse running CCTV.
Questions: - what is the measurement precision?
- do you need calibration? How do you do it in production?
- what it is the root problem you are trying to solve?
- what is your hypothesis about your solution- quantitatively?
2) Not necessarily. Models like MapAnything/MoGe predict calibration params directly and GeoCalib is good for distortion coefficients. We still calibrate manually on-site, but mainly to validate these models actually hold up in real warehouses and collect our own calibration training data. We are confident the future is calibration-free.
3) carriers lose money every day because shippers understate dimensions and LTL is priced by volume. Every understated shipment is lost revenue. Thats the wedge we sre going after.