The LUMME kCalculator is a machine learning smartwatch app that you can train like a puppy to passively sense when you’re eating from your hand motions and scientifically measure your energy intake. Once trained, detection will occur as long as you’re eating for at least a few minutes at a time and you’re bringing food to your mouth. You can eat as you normally would with a fork, spoon or hands.
First ensure that you are syncing your motion sensor data. Your main meal diary screen will provide a ‘Last data synced’ timestamp. If it’s blank, please follow the instructions for data syncing. Otherwise, here are a few suggestions:
1. Check Find Missing Meals eating detection under different sensitivity.
2. More Training data is required. Go to Settings > Training.
3. Take at least a couple bites a minute for at least 3 minutes
4. Bring food to your mouth and not vice-versa (hand movement is essential)
5. Watch should be on the eating hand
6. Open up the app sometime after you eat for faster data processing
Let us know what you’re experiencing, and we’ll be able to provide some additional guidance.
Here are the most common ways to resolve this problem:
1. On your watch, please verify that the kCalculator complication is visible on the watch face you use. The complication is essential for data transfer. If you missed this during the watch configuration, you’ll find some guidance here.
2. Launch the kCalculator app on your phone and watch. Please ensure that you do not remove the kCalculator app using the App Switcher UI on your phone. This kills the app and prevents it from detecting eating.
3. On your phone, open Watch app > Privacy > Enable Fitness Tracking (if not enabled already).
4. On your phone, open Watch app > General > Background App Refresh > Enabled background app refresh (if not enabled already).
5. Restart your iPhone and then your Apple Watch.
6. If there are any other issues, the phone app will give you alert messages upon app launch.
Contact us at email@example.com for additional assistance.
There are 4 key ways in which the kCalculator learns:
1. Deleting false detections.
2. Confirming correct detections by assigning those sessions as breakfast, lunch, dinner or snack.
3. Running a Training Session for at least 10 bites.
4. Retrieving missed detections from Find Missing Meals
Similar to a voice-to-text dictation app that learns the subtleties of your voice, “training” the kCalculator app to learn the subtleties of your eating gestures, even for a few minutes, can result in dramatic improvements in eating detection. We recommend using the Training Session module in the Settings menu at the start of using the app, and as often as you like. Try training in the key places where you sit (or stand) to eat.
It’s easy. Just run it for 10 bites at your next meal. Following the instructions provided, simply take a bite of food, tap the button with your off-hand, and repeat. The effects of a training session will kick in the following day.
Your Meal Verification Score is the leading indicator of your app’s level of personalization. When you CONFIRM an eating detection or DELETE it, you’re actually using machine learning to train your app to distinguish between your TRUE and False eating gestures. Aim to keep this high and watch your detection accuracy and calorie estimates improve with each successive day of use.
If your app ‘fails’ to detect an eating session, there’s a good chance you’ll find it under Find Missing Meals (FMM). FMM allows you to see eating detections under different Eating Detection Sensitivity settings.
Your default eating detection sensitivity is set to ‘LESS Detections’, but you can view eating detections sensed under the ‘MORE‘ or ‘MOST‘ settings. Once a detection is Accepted (Select, Save & Exit), you can edit or delete it within your Meal Diary. You can also update your Default detection setting on this page.
If you do NOT find your missing meal here, you can use the Add Calories Manually option. This might also be a good indication that you should conduct a Training Session where the meal was missed.
Your kCalculator filters over 30K gestures in a day to identify the 100+ times you actually put food to your mouth. So some false detections will occur. By deleting any falsely detected eating sessions in your diary, you’ll teach your app to do better.
The kCalculator is continuously filtering through tens of thousands of gestures to identify the times when you are eating. When it identifies an eating session, it will create a meal record in your diary and ask for confirmation on your watch or phone.
Apple’s iOS controls when and where eating detection notifications are issued. Typically, this can range from 15 – 60 min after you have eaten. The variance is due to background processing factors dictated by the iOS environment. Some of the many factors include battery life, app usage and app slider visibility.
If you miss a notification, the kCalculator will still create an unconfirmed ‘Eating Detected’ meal record.
Open up your kCalculator phone app for about 30 seconds sometime after you’ve finished eating (wait at least 5 minutes after eating). This will ‘force’ the synchronization of your sensor data for processing.
In addition to seeing your data faster, iOS will start to give your app better background processing priority, which creates a virtuous cycle.
What happens if you don’t open the app?
iOS will sync your data as determined by its algorithms. We observe huge variability from once per hour to many hours.
Also, as the time lengthens between synchronizations, so does the build-up of your data, sometimes creating longer delays or requiring multiple synchronization routines to catch up.
If you have cellular or wi-fi connectivity issues, this can also slow things down, but you’ll otherwise start to see the benefits of reviewing your meal data a few times a day. This will also make it much easier to verify your meal detections and delete any false detections, which improves personalization!
Does Siri correctly spell every word that you dictate through the microphone? Not likely, but it continues to improve. Similarly, the kCalculator is designed to naturally improve based on your inputs. Each input from you fine-tunes the app for better detection accuracy and calorie estimates. The Training tool within the app adds a heavy dose of personalization to your app. So does Find Missing Meals, when you import a missed detection. Basic Meal Verification is also essential for personalization. The more it learns, and the more training it receives, the better it will detect your eating. For some perspective, the app may sift through upwards of ninety-thousand hand gestures in day to identify just 100+ instances where you put food to your mouth!
kCalculator uses scientifically-validated bite-based estimates, which use an objective caloric proxy for your bites according to various profile details, including your gender, age, height and current weight. This handles the volume of food that you eat (your portion sizes), which is the major driver of your daily calories. To account for variances in energy density associated with what you eat, the kCalculator gives you the option to refine your estimate using the Calorie Type slider.
Your food choices will always matter to your health and impact your calories. It’s also true that the major driver of your daily calories comes down to your portion sizes – simply HOW MUCH YOU EAT. kCalculator measures this volume of food from your bites. The vast majority of variance in energy density associated with WHAT YOU EAT can be captured with one simple question regarding your ‘Meal Type’. kCalculator defaults to a Meal Type of Mixed High and Low Calorie Foods, but you can use the simple slider in your Meal Diary to adjust your Meal Type to High Calorie Foods or Low Calorie Foods.
Calorie Types provide a simple and effective way to categorize your meals and snacks according to the energy density of the foods you eat. For simplicity, kCalculator defaults your meals and snacks to the Mix of High and Low Calorie Foods calorie type. If you’re eating mainly Low Calorie Foods (salad, fish, fruit, vegetables) or High Calorie Foods (burgers, pizza, sweets), use the simple slider in your Meal Diary to select these alternative options.
Take a look at the tables below to get a better understanding of the Meal Types.
Calories are a measure of energy intake and can be used in conjunction with a calorie goal or as a point of reference to understand your eating habits better. Just keep in mind that your food choices still matter to good health. Be sure to check out the “What to Know About Counting Calories to Lose Weight“. It’s packed with great information.
Beverages can range from zero calories (water) to hundreds of calories (meal replacement shakes). If you consume caloric beverages such as sugary drinks, smoothies, alcoholic beverages or meal replacement shakes, your Meal Diary provides a simple way to add these calories. It’s often easy to find the calorie information for these beverages on the nutrition label, just be sure to pay close attention to the portion size that you consumed.
kCalculator aims to simplify calorie tracking by providing a balance between accuracy, reliability and effort. The most precise calorie measurement techniques, such as weighing individual food items and looking up their energy composition, are also the most cumbersome and impractical. Even calorie tracking apps require that you correctly estimate your portion sizes. With minimal effort, kCalculator offers a useful and reliable daily measure of energy intake that can be more accurate than human-based estimates even with the aid of calorie information. Keep in mind that the accuracy will improve over time and be quickly enhanced through Training Sessions.
LUMME’s kCalculator uses machine learning that improves over time based on your responses to eating detection notifications and as more sensory data is fed into your personalized AI model. The app’s Training Session function can be used to learn your more unique eating gestures to speed up this process. We encourage you to conduct a short Training Session to personalize your app right from the start and focus on the particular foods and/or plate settings that it has trouble detecting.
Your Meal Diary is your record of detected eating episodes, as well as a place to manually enter calories if necessary. A meal diary record is automatically recorded when eating is detected. The Meal Diary is an easy place to review the times that you eat, make any adjustments.
Your Weight Tracker is a simple tool to keep track of your weight. If you’re using the kCalculator for a self-help weight loss program, entering your weight on a regular basis can be helpful to track progress and keep your calorie estimates as accurate as possible.
Mindful eating is the practice of eating with purposeful awareness. Volumes have been written about the importance of mindful eating, its benefits and how to do it, but the Mindful Eating tab guides you through the process. Use the Mindful Eating tab as a 20-minute meal companion to help you master the skills of pace, bite-size, and greater enjoyment of your food.
A qualitatively better approach requires a few new rules. Once you know them, a powerful new system for change is yours. Soon you’ll be unlocking deeper insights into your eating behavior, predictions to stay ahead of problem areas, and newfound strength and control to achieve your goals. Here’s how to sharpen your new personal health tool.
If you’re right-handed, you probably wear your watch on your left hand. And, if you’re left-handed you probably wear your watch on your right hand. Making the transition to the hand you use for eating will be a snap for some, and can take a few weeks for others. But, here’s the secret. No one will ever notice the switch. And, the body and brain will habituate to the new physical sensation. To avoid any unnecessary stress or discomfort, and enable the habituation process, try a soft, lightweight strap to allow for micro-adjustments. To stay comfortable, you can also build up the time you spend wearing your watch each day.
The kCalculator looks for eating sessions that last at least a 2-3 minutes. You should also be bringing food to your mouth, not vice versa.
Detecting eating while you’re moving around or driving is a lot trickier due to the additional ‘noise’ added to your motion sensor data.
We created the Training tool to inject a heavy dose of personalization into your app any time you need it, but especially on Day 1. Look for the option under Settings. To get started, simply “record” 10 bites of food at your next meal. Eat whatever, wherever, and however you normally would. Generally, your app will personalize faster and detect more accurately with more training so give your app some variations. Missed detections are a good signal your app needs training. You can train with certain foods, dining tables or seating variations where it fails to detect. Training benefits will kick in the following day.
In addition to training your app to recognize the subtleties of your eating gestures, verifying your eating detections also personalizes your app. The dials on your Meal Diary screen will show you how you’re doing. You can assign a detection to Breakfast, Lunch, Dinner or Snack. There’s also a ‘Not Sure’ option that keeps the meal record, but removes it from your calorie estimates. You can also delete the detection record.
Your kCalculator sifts through 30,000 – 90,000 gestures in a day to identify the 100+ times you actually put food to your mouth. So, there will be false detections. The number of false detections will vary based on your lifestyle and the uniqueness of your gestures, but the only way to weed out false detections is by deleting them. Your app will learn from these actions and will gradually eliminate these from your recognized eating gestures. Delete your false detections by swiping left on the meal record.
Here’s the deal. Apple (iOS/WatchOS) controls which of your apps get priority and when. It can make a big difference in determining when, where and even if you receive an eating detection notification. The only way give you app higher prioritization so you can see your eating detections faster is to manually sync your data. Simply open up your phone app and further your watch app for about 30 seconds after you’ve finished eating (wait at least 5-10 minutes). This will load your most recent detected eating sessions. If you have cellular or wi-fi connectivity issues, this can slow things down, but you’ll otherwise start to see the benefits.
Due to the high caloric variability of liquids, the kCalculator will do its best to exclude all sips (drinking gestures) from your calorie estimates. If it does pick up a drinking event like your morning coffee, DELETE it. The ‘+‘ button in the meal diary makes it easy to manually log any caloric beverages or missed events you’d like to record.
The kCalculator is a tool that can feel blunt at first. You’re likely to detect too much or too little, and calorie estimates will be off. It needs sharpening through training personalization and meal verification data. Accuracy isn’t a question of if — it’s about how much personalization you require to capture the vast majority of your eating. Less consistent eating habits may require more personalization data. Contact us if you’re not seeing dramatic improvements through your personalization.
If you’re not detecting any eating on Day 1, regardless of any Training, we should do some troubleshooting right away. Send us a quick note at firstname.lastname@example.org. It will be helpful if you could identify a specific eating event or two (approx time/duration) for us to investigate.
Before facial recognition was introduced to the iPhone, we registered our fingerprints. It required 5-10 different position variations to get the complete print. The kCalculator needs to do something similar to properly distinguish your unique eating gestures from the 50K+ observed each day.
Follow these 3 essential tips to dial in the accuracy of your detections and calorie estimates.
Delete your false detections and confirm the correct ones by assigning them to Breakfast, Lunch, Dinner, Snack or even Not Sure. The app will learn from this.
A single training session improves detection accuracy by 34% on average. It’s highly recommended that you run a Training session on Day 1. Consider the handful of places where you eat (dining table, couch, standing in kitchen, office desk…) as an opportunity to train and complete your gesture profile.
Training Sessions are easy, non-disruptive, and incorporate your normal eating in just 10 bites. Go to Settings/Training.
Find Missing Meals allows you to located ‘missed’ detections under different detection sensitivity levels. It’s not just useful, it’s powerful. Any entries that you import into your meal diary will be used to further improve detection.
Foods in the “low” category are higher in water and fiber content. Foods in the “high” category are higher in sugar and/or fat content. Foods in the “medium” category tend to contain a mix.
Low Energy Density Categories
Medium Energy Density Categories
High Energy Density Categories
Research shows that nearly any dietary approach can work for weight loss if you stick to it and reduce the number of calories that you consume, and the kCalculator works well with all of them! See below for a day’s worth of example meals and snacks from a variety of healthy dietary approaches that are commonly used for weight loss. Each example totals to about 1,500 calories per day, which is a commonly used calorie goal for weight loss.
MEAL 1: 1/2 avocado + 1 hard boiled egg + 1 medium orange = 331 kCal
MEAL 2: 1/2 can of light tuna in olive oil + 1/2 cup lettuce + 1 tbsp of mayo + 1 medium apple + 1 oz of cheddar = 365 kCal
MEAL 3: 4 oz grilled chicken breast + 1 cup of spaghetti squash + 2 tablespoon pesto sauce + 1/4 cup of shredded cheese = 480 kCal
SNACK 1: 1/2 cup blueberries + 1/2 cup of nonfat yogurt + 2 tbsp unsweetened shredded coconut = 200 kCal
SNACK 2: 1/4 cup of hummus + 5 medium celery stalks = 125 kCal
TOTAL CALORIES = 1501 kCal
MEAL 1: 1/2 cup oatmeal + 1/2 cup of blueberries + 1/2 cup almond milk = 230 kCal
MEAL 2: 2 slices of whole grain bread + 2 oz of turkey deli meat + lettuce + sliced tomato + 1 slice of provolone + medium apple = 500 kCal
MEAL 3: 4 turkey meatballs + 1/2 cup of whole grain pasta + 1/4 cup marinara sauce + 1/4 cup fat free shredded mozzarella = 500 kCal
SNACK 1: 1/2 cup nonfat cottage cheese + 1/2 cup pineapple = 130 kCal
SNACK 2: 1/2 cup of nonfat yogurt + 1/2 cup sliced strawberries = 130 kCal
TOTAL CALORIES = 1490 kCal
MEAL 1: 1 cup of nonfat plain Greek yogurt + 1/2 cup of blackberries + 3 tbsp chopped walnuts = 292 kCal
MEAL 2: 10 medium shrimps + 1 cup side salad ( mixed veggies) + 1/2 avocado = 390 kCal
MEAL 3: 6 oz of salmon +
1 tbsp olive oil + 1/2 sweet potato + 1/2 cup broccoli = 550 kCal
SNACK 1: 22 dry roasted almonds + 1 clementine = 200 kCal
SNACK 2: 1 medium pepper + 2 tablespoons hummus = 75 kCal
TOTAL CALORIES = 1507 kCal
MEAL 1: 2 eggs scrambled + 1 slice whole grain toast = 240 kCal
MEAL 2: 1/2 cup cooked tofu + 1/2 cup of lettuce + 1/4 cup black beans + 1 red pepper chopped + 2 tbsp Balsamic vinaigrette = 448 kCal
MEAL 3: 2 hard taco shells + 1/2 cup of refried beans + 1/4 cup shredded nacho cheese + 1/4 tomato + 1/2 cup shredded lettuce + 1 tbsp sour cream = 483 kCal
SNACK 1: 1/2 cup of roasted chickpeas = 135 kCal
SNACK 2: 1/4 cup cashews + plum = 220 kCal
TOTAL CALORIES = 1526 kCal