1- Comparison of Detectrons output with our own labels:

[m-kashani]

First figure out your top 20 images and save them accordingly.

• [ ] Question -> Images with exactly 10 objects in them?

---

To read and understand the pipeline:

• [x] Digging into Detectron Part1:

• [ ] https://medium.com/@hirotoschwert/digging-into-detectron-2-part-2-dd6e8b0526e

• [ ] https://medium.com/@hirotoschwert/digging-into-detectron-2-part-3-1ecc27efc0b2

• [ ] Proposal Generator?

• [ ] Box Head.

---

extra Amir

• [ ] https://gilberttanner.com/blog/detectron-2-object-detection-with-pytorch

[m-kashani]

├─layers <- custom layers e.g. deformable conv. https://github.com/facebookresearch/detectron2/tree/master/detectron2/layers

[m-kashani]

Meta Architecture:
GeneralizedRCNN (meta_arch/rcnn.py)
which has:
Backbone Network:
FPN (backbone/fpn.py)
└ ResNet (backbone/resnet.py)
Region Proposal Network:
RPN(proposal_generator/rpn.py)
├ StandardRPNHead (proposal_generator/rpn.py)
└ RPNOutput (proposal_generator/rpn_outputs.py)
ROI Heads (Box Head):
StandardROIHeads (roi_heads/roi_heads.py)
├ ROIPooler (poolers.py)
├ FastRCNNConvFCHead (roi_heads/box_heads.py)
├ FastRCNNOutputLayers (roi_heads/fast_rcnn.py)
└ FastRCNNOutputs (roi_heads/fast_rcnn.py)

[m-kashani]

look at the outputs. See https://detectron2.readthedocs.io/tutorials/models.html#model-output-format for specification

print(outputs["instances"].pred_classes) print(outputs["instances"].pred_boxes)