Incomplete chain output in some cases

[rguilmont]

Hi !
I noticed an issue with /chain endpoint, with an incomplete output in the following case :

Given this request

{
    "ressource": "predict",
    "calls": [
        {
            "service": "face_detection",
            "data": [
                "https://www.vishopper.com/images/products/maxxmax/PE/10635_a-large-crowd-standing-and-listening-to-a-concert.jpg"
            ],
            "parameters": {
                "input": {
                    "connector": "image",
                    "keep_orig": true
                },
                "output": {
                    "confidence_threshold": 0.7,
                    "bbox": true
                },
                "mllib": {
                    "gpu": true,
                    "net": {
                        "test_batch_size": 1
                    }
                }
            },
            "parent_id": ""
        },
        {
            "id": "face_crop",
            "action": {
                "type": "crop"
            }
        },
        {
            "service": "age",
            "parameters": {
                "input": {
                    "connector": "image",
                    "keep_orig": true
                },
                "output": {
                    "best": 100,
                    "confidence_threshold": 0.001
                },
                "mllib": {
                    "gpu": true,
                    "net": {
                        "test_batch_size": 1
                    }
                }
            },
            "parent_id": "face_crop"
        }
    ]
}

if no faces are detected, then the final output will be like :

{"status":{"code":200,"msg":"OK"},"head":{"method":"/chain","time":1132.0},"body":{}}

But I expected at least to have the list of images in the response, like /predict ressource, even if there's in the end classes is empty, like for instance :

{
    "status": {
        "code": 200,
        "msg": "OK"
    },
    "head": {
        "method": "/chain",
        "time": 1132.0
    },
    "body": {
        "predictions": [
            {
                "uri": "https://www.vishopper.com/images/products/maxxmax/PE/10635_a-large-crowd-standing-and-listening-to-a-concert.jpg",
                "classes": [
                ]
            }
        ]
    }
}

Configuration

DeepDetect version : commit 2b61a52ce132202bf62283eccac1b40b48398dd8, on Docker ( ubuntu 16.04 based image )
GPU : Nvidia T4 with Cuda 10.1