Detection

UP supports the whole pipline of training, deploying, and interfering;

Codes

Configs

It contains the illustration of common configs and deploying.

Repos

Dataset related modules

1. Dataset types:

• coco

• custom

• lvis

2. The type of datasets and the config is as followed:

dataset:
  type: coco   # [coco, custom, lvis]
  kwargs:
    meta_file: coco/annotations/instances_train2017.json
    image_reader:
      type: fs_opencv
      kwargs:
        image_dir: coco/train2017
        color_mode: RGB
    transformer: [*stitch_expand, *crop, *flip, *color_jitter, *fix_output_resize,*to_tensor, *normalize]

3. Data augmentation:

• StitchExpand:

expand: &stitch_expand
  type: stitch_expand
  kwargs:
    expand_ratios: 2.0
    expand_prob: 0.5

• ImageCrop:

crop: &crop
  type: crop
  kwargs:
    means: [123.675, 116.280, 103.530]
    scale: 1024
    crop_prob: 0.5

Loss fuction

UP supports focal loss, iou loss, and smooth l1 loss.

focal loss:

• sigmoid_focal_loss

loss:
  type: sigmoid_focal_loss
  kwargs:
    num_classes: *num_classes
    alpha: 0.25
    gamma: 2.0
    init_prior: 0.01

• quality_focal_loss

loss:
  type: quality_focal_loss
  kwargs:
    gamma: 2.0
    init_prior: 0.01

iou loss:

loss:
  type: iou_loss
  kwargs:
    loss_type: iou  # type: ['iou', 'giou', 'diou', 'ciou', 'linear_iou', 'square_iou']

smooth_l1_loss:

loss:
  type: smooth_l1_loss
  kwargs:
    sigma: 3.0

compose_loc_loss:

loss:
  type: compose_loc_loss
  kwargs:
    loss_cfg:
      - type: iou_loss
        kwargs:
          loss_type: giou
          loss_weight: 1.0
      - type: l1_loss
        kwargs:
          loss_weight: 1.0

Deploying modules

‘ToKestrel’ is needed when models are transformed to kestrel models as followed.

1. Add ‘class_names’ to dataset.train, take COCO as an example:

class_names: &class_names [
"__background__", "person", "bicycle", "car", "motorcycle", "airplane",
"bus", "train", "truck", "boat", "traffic light", "fire hydrant",
"stop sign", "parking meter", "bench", "bird", "cat", "dog", "horse",
"sheep", "cow", "elephant", "bear", "zebra", "giraffe", "backpack",
"umbrella", "handbag", "tie", "suitcase", "frisbee", "skis", "snowboard",
"sports ball", "kite", "baseball bat", "baseball glove", "skateboard",
"surfboard", "tennis racket", "bottle", "wine glass", "cup", "fork",
"knife", "spoon", "bowl", "banana", "apple", "sandwich", "orange",
"broccoli", "carrot", "hot dog", "pizza", "donut", "cake", "chair",
"couch", "potted plant", "bed", "dining table", "toilet", "tv", "laptop",
"mouse", "remote", "keyboard", "cell phone", "microwave", "oven",
"toaster", "sink", "refrigerator", "book", "clock", "vase", "scissors",
"teddy bear", "hair drier", "toothbrush"
]

dataset:
  train:
    dataset:
      type: coco
      kwargs:
        meta_file: coco/annotations/instances_train2017.json
        class_names: *class_names
        image_reader:
          type: fs_opencv
          kwargs:
            image_dir: coco/train2017
            color_mode: RGB
        transformer: [*flip, *train_resize, *to_tensor, *normalize]

2. Add ‘to_kestrel’ to config:

to_kestrel:
  detector: FASTER-RCNN  # choices [RETINANET, RFCN, FASTER-RCNN]
  model_name: model  # prefix of tar-model filename and model_name in meta.json
  default_confidence_thresh: 0.3
  plugin: harpy   # choices = [essos, harpy, sphinx]
  version: 1.0.0
  resize_hw: 640x1024
  kestrel_config:   # 可选
    -
      # harpy
      confidence_thresh: 0.3
      # essos
      thresh: 0.3
      id: 37017
      label: face
      filter_h: 0
      filter_w: 0