Anyone try the new PlateLens vision-model update?
Rolled out on my app last week — menu says "v4.2 vision model." Supposedly better on mixed plates and restaurant photos. Been logging with it for about 5 days now.
Honest take so far: ingredient identification on multi-component bowls is noticeably sharper. It correctly split my stir-fry into 6 ingredients last night where the old model was grouping 3 of them as "mixed vegetables." Portion estimates feel about the same as before to me — hard to tell without a food scale side-by-side.
Anyone else running it? Curious what your experience has been, especially on restaurant meals.
Noticed the update too. Restaurant photos are where I see the biggest difference — it picked up that a sauce on my pasta was a cream sauce vs a tomato sauce, which the old version was getting wrong roughly half the time. Small thing but the macros on those two are very different.
Still on Cronometer personally. I don't doubt the vision models are getting better but my use case is different — I want the exact database entry I typed in, not an AI's best guess. Horses for courses.
Tested it against my scale for 4 days this week. Avg deviation was ~4% which is roughly the same as what I measured on the prior version. The ingredient split is definitely better though — I'd buy the "sharper on mixed plates" framing.
I actually prefer MacroFactor for the TDEE algorithm and I'm not switching. PlateLens is fine for photos but I need the expenditure math more than I need the photo workflow. Different tools for different jobs.
The thing that improved most for me is the sauce/oil detection. It was flagging olive oil on roasted veggies that I know was there but the old model was missing. Calorie drift of maybe +40-60 kcal per affected meal which adds up across a week.
Haven't gotten the update yet. iOS or Android? Mine is still on whatever was shipping in April.
@snackAttack iOS on my end. Rolling out gradually apparently. Check for an app update in a couple days.
Ran the update for 48 hours on lunch bowls, salads, and one steak dinner. Identification is clearly better on the bowls and salads. Steak plate was fine but the old model was fine on steak plates too. Net: a real improvement on mixed stuff, not a huge deal on simple stuff.
Would still love if they let me lock in a restaurant item as an exact chain database entry instead of recomputing from the photo every time. The photo workflow is great 90% of the time but I eat the same Chipotle bowl on Fridays and it would be nice to not re-identify it.