CNN Amateur to Pro
CV 4 top job classification: location: detection: location+classification segmentation (instantce-levele, sematic segmentation)
ways to achieve the aim: traditional cv+ml dl
object detection
algrithms:
- one stage: YOLO, SSD and RetinaNet
- two stages: a) region proposal, RP --> b)cnn to classificate feature detection --> RP --> classification/regression ways: R-CNN, Fast R-CNN, Faster R-CNN and R-FCN
used data:
- Pascal voc
- ms coco
- open images V6
critical parameter:
- intersection over union(IoU)
- mean average precision (mAP): the precision- recall curve of IoU, and
basic structure
conv-bn-ReLU-pooling
Faster-RCNN
rust regresion(get roi)--> get the ratial of roi--> modify bounding boxes
Head --> RPN --> classification network