[r0bbbo]

This is what Microsoft Cognitive Services saw:

Objects [ { "rectangle": { "x": 346, "y": 321, "w": 333, "h": 417 }, "object": "mammal", "parent": { "object": "animal", "confidence": 0.719 }, "confidence": 0.716 } ]

Tags [ { "name": "wall", "confidence": 0.9889654 }, { "name": "indoor", "confidence": 0.9775573 }, { "name": "doll", "confidence": 0.9775573 }, { "name": "person", "confidence": 0.966689169 }, { "name": "toy", "confidence": 0.5887871 }, { "name": "collection", "confidence": 0.535430968 }, { "name": "bear", "confidence": 0.432904869 }, { "name": "christmas", "confidence": 0.351226926 }, { "name": "animal", "confidence": 0.350886434 }, { "name": "family", "confidence": 0.3459218 } ]

Description { "tags": [ "indoor", "table", "sitting", "food", "dog", "woman", "black", "bear", "holding", "standing", "man", "room", "people", "bed", "group", "stuffed", "red", "kitchen" ], "captions": [ { "text": "a group of stuffed animals sitting on top of a table", "confidence": 0.764803648 } ] }

[firethief]

Now that I look at it again, there definitely is a wall in the back left. MCS succeeded where I failed because I guess I have an implicit "walls aren't interesting" filter that threw out the only positively identifiable object.

[patrickmcnamara]

Here's the top ten things Amazon Rekognition saw:

Furniture - 94.8 %, Apparel - 93.8 %, Clothing - 93.8 %, Home Decor - 93.6 %, Table - 85.3 %, Pet - 77.9 %, Mammal - 77.9 %, Cat - 77.9 %, Animal - 77.9 %, Toy - 67.9 %

It's surprisingly confident.

[tabtab]

GobbledyGook: confidence = 99.9937%

[jschwartzi]

better than I could do. I can't make hide nor hair of it.

[dhritzkiv]

Is it really "better" if the image is simulated nonsense? Better would be conceding that nothing is truly decipherable.