I tested the Best Calorie Counter GPT today to see how accurate it really is by taking 4 meals and doing two separate tests for each recipe.
The first test was done on each meal by taking a photo of the meal and asking for the Calories and Protein totals.
The second test was providing a prompt with the ingredients in a normal sentence and asking for the same.
I was surprised with the outcome.
Taking the simple sample of 4 separate meals, for which I had all the precise ingredients, as well as the exact Calories and Protein breakdown for each, here are the results.
Baseline Meal 1. Curry Noodle Soup with Coconut and Udon Noodles. Cal: 672, Protein 18g, Carbs 55g
Best Calorie Counter Photo Method: Cal:975, Protein 24g, Carbs, 61g. This measurement was too high. It estimated the peanuts as a quarter cup, which was too much by 100g Calories. Also it overestimated the amount of noodles by 100g which explains the high Calorie count. It surprisingly got the ingredients spot on.
Best Calorie Counter Prompt method: Calories 657g and Protein18g. This was right in the ballpark.
Baseline Meal 2. Rigatoni with beef mince and aubergine cream soup: Calories 866, Protein 34g
Best Calorie Counter Photo Method: Calories775, Protein 44g. This measurement was close enough. The visible ingredients were almost with correct except that it mistook Rucola for Arugula. From the image the 100ml cooking cream was not picked up by the AI and neither was the olive oil which compensates for the low Calorie count. Understandable.
Best Calorie Counter Prompt method: Calories 884g and Protein 40g. This was right in the ballpark again.
Baseline Meal 3. Vegetarian quesadillas with spicy cheese blocks: Calories 675, Protein 24g
Best Calorie Counter Photo Method: Calories:677, Protein 46g. This measurement was really close. The full grain tortillas were picked as normal flour tortillas, and the cheese was overr estimated a bit. It mistakenly picked up the courgette as chicken, which accounts for the high protein number. Overall i was impressed with this one.
Best Calorie Counter Prompt method: Calories 671 and Protein 22.3g. This was right in the ballpark again. Very accurate.
Baseline Meal 4. Farfalle pasta with chicken in cream sauce: Calories 818, Protein 41g
Best Calorie Counter Photo Method: Calories 646, Protein 48g. This calories were low on this one. The full grain tortillas were picked as normal flour tortillas, and the cheese was overr estimated a bit. It mistakenly picked up the courgette as chicken, which accounts for the high protein number. Overall i was impressed with this one.
Best Calorie Counter Prompting method: Calories 822 and Protein 40g. This was right in the ballpark again. Very accurate.
What can I deduce from the above?
The Best Calorie Counter is really accurate if you have the precise weight or volume of the ingredients. It makes Calorie Counting really easy, without having to look up each item individually and then add up the individual values to the total. Simply provide the ingredients in a normal sentence and all is taken care for you, providing accurate individual totals and the grand total of Calories and Protein values for your entire meal.
If you do not know the exact weights or volumes, the image route is close enough, but has some caveats. The image recognition sometimes misses some hidden ingredients like olive oil and cooking cream, for instance. It also can mistake things like courgette for chicken.
In my tests this was easy to rectify by simply telling it that it is a vegetarian dish, or that theres also creamy sauce or olive oil in the dish and then ask it to recalculate. You will get almost perfect results this way.
This specific method is ideal if you go to friends or a restaurant for lunch or dinner, but still want to keep count of the Calories you ingested. Simply take a picture of whatever you eat and when you're at home simply have the AI analyse it for you, and record it in your favourite app like normal.
The meal analysis is really good, will save you a ton of time and with a bit of common sense it will provide you with really accurate estimations of your meals away from your scale.
The GPT is here and free to use by anyone. (you do need a GPT plus account though)
https://chat.openai.com/g/g-ASM8P8VNI-best-calorie-counter
Let me know if anyone here have done any similar testing and what the outcomes or results were.
The first test was done on each meal by taking a photo of the meal and asking for the Calories and Protein totals.
The second test was providing a prompt with the ingredients in a normal sentence and asking for the same.
I was surprised with the outcome.
Taking the simple sample of 4 separate meals, for which I had all the precise ingredients, as well as the exact Calories and Protein breakdown for each, here are the results.
Baseline Meal 1. Curry Noodle Soup with Coconut and Udon Noodles. Cal: 672, Protein 18g, Carbs 55g
Best Calorie Counter Photo Method: Cal:975, Protein 24g, Carbs, 61g. This measurement was too high. It estimated the peanuts as a quarter cup, which was too much by 100g Calories. Also it overestimated the amount of noodles by 100g which explains the high Calorie count. It surprisingly got the ingredients spot on.
Best Calorie Counter Prompt method: Calories 657g and Protein18g. This was right in the ballpark.
Baseline Meal 2. Rigatoni with beef mince and aubergine cream soup: Calories 866, Protein 34g
Best Calorie Counter Photo Method: Calories775, Protein 44g. This measurement was close enough. The visible ingredients were almost with correct except that it mistook Rucola for Arugula. From the image the 100ml cooking cream was not picked up by the AI and neither was the olive oil which compensates for the low Calorie count. Understandable.
Best Calorie Counter Prompt method: Calories 884g and Protein 40g. This was right in the ballpark again.
Baseline Meal 3. Vegetarian quesadillas with spicy cheese blocks: Calories 675, Protein 24g
Best Calorie Counter Photo Method: Calories:677, Protein 46g. This measurement was really close. The full grain tortillas were picked as normal flour tortillas, and the cheese was overr estimated a bit. It mistakenly picked up the courgette as chicken, which accounts for the high protein number. Overall i was impressed with this one.
Best Calorie Counter Prompt method: Calories 671 and Protein 22.3g. This was right in the ballpark again. Very accurate.
Baseline Meal 4. Farfalle pasta with chicken in cream sauce: Calories 818, Protein 41g
Best Calorie Counter Photo Method: Calories 646, Protein 48g. This calories were low on this one. The full grain tortillas were picked as normal flour tortillas, and the cheese was overr estimated a bit. It mistakenly picked up the courgette as chicken, which accounts for the high protein number. Overall i was impressed with this one.
Best Calorie Counter Prompting method: Calories 822 and Protein 40g. This was right in the ballpark again. Very accurate.
What can I deduce from the above?
The Best Calorie Counter is really accurate if you have the precise weight or volume of the ingredients. It makes Calorie Counting really easy, without having to look up each item individually and then add up the individual values to the total. Simply provide the ingredients in a normal sentence and all is taken care for you, providing accurate individual totals and the grand total of Calories and Protein values for your entire meal.
If you do not know the exact weights or volumes, the image route is close enough, but has some caveats. The image recognition sometimes misses some hidden ingredients like olive oil and cooking cream, for instance. It also can mistake things like courgette for chicken.
In my tests this was easy to rectify by simply telling it that it is a vegetarian dish, or that theres also creamy sauce or olive oil in the dish and then ask it to recalculate. You will get almost perfect results this way.
This specific method is ideal if you go to friends or a restaurant for lunch or dinner, but still want to keep count of the Calories you ingested. Simply take a picture of whatever you eat and when you're at home simply have the AI analyse it for you, and record it in your favourite app like normal.
The meal analysis is really good, will save you a ton of time and with a bit of common sense it will provide you with really accurate estimations of your meals away from your scale.
The GPT is here and free to use by anyone. (you do need a GPT plus account though)
https://chat.openai.com/g/g-ASM8P8VNI-best-calorie-counter
Let me know if anyone here have done any similar testing and what the outcomes or results were.