mukeshkdangi / custom-ai

work with the cognitive service provided by Microsoft Azure to build image classifier

customeai

Let first upload some images of the two different card Jeep and Mercedes

Then we train and build a model out these uploaded images:

POST https://southcentralus.api.cognitive.microsoft.com/customvision/v2.0/Prediction/{PID}/url?iterationId={iterationId} HTTP/1.1 Host: southcentralus.api.cognitive.microsoft.com Prediction-Key: {PKEY} Content-Type: application/json

{ "Url": "https://di-uploads-pod11.dealerinspire.com/raylaethemchryslerdodgejeep/uploads/2017/09/2018-Jeep-Compass-Hero.png" }

Let's test our model with the one JEEP image see how accurate our model is

apim-request-id: {id} Strict-Transport-Security: max-age=31536000; includeSubDomains; preload x-content-type-options: nosniff Date: Mon, 25 Jun 2018 21:01:34 GMT Content-Length: 389 Content-Type: application/json; charset=utf-8

{ "id": "{id}", "project": "{pid}", "iteration": "{iteration}", "created": "2018-06-25T21:01:34.487003Z", "predictions": [{ "probability": 0.468212515, "tagId": "b2af220e-a4f4-43b7-9c98-c13a6c9f8ca1", "tagName": "Mercedes" }, { "probability": 0.367403924, "tagId": "8e958793-dfd4-40c5-92ee-b861c7005b5e", "tagName": "Jeep" }] }

So our model is not si good as it's saying the new image os 47% Mercedes and 36% channces that it could be Jeep

Now we'll uploaded more images so that our system can understand the differnces b/t these 2 cars very well

Once we are doing the training phase 2, we test same image against our model then see the result :

Http Response : apim-request-id: {id} Strict-Transport-Security: max-age=31536000; includeSubDomains; preload x-content-type-options: nosniff Date: Mon, 25 Jun 2018 20:59:48 GMT Content-Length: 388 Content-Type: application/json; charset=utf-8

{ "id": "{id}", "project": "{pid}", "iteration": "{iteration2}", "created": "2018-06-25T20:59:48.4573752Z", "predictions": [{ "probability": 0.9310171, "tagId": "8e958793-dfd4-40c5-92ee-b861c7005b5e", "tagName": "Jeep" }, { "probability": 0.166567564, "tagId": "b2af220e-a4f4-43b7-9c98-c13a6c9f8ca1", "tagName": "Mercedes" }] }

So our model is not si good as it's saying the new image os 93% Mercedes and only 16% channces that it could be Jeep