Spacecraft Detection in Images Using Machine Learning

Authors: - Nick Duggan - Robert Mancini - Luke Spinosa

Running the Code

Training

All training occurs in the scripts/main.py file. To train the model, you must first be inside of the scripts directory bash cd scripts then run the file with the following command: bash python main.py A progress bar will display while training is occuring indicating the current batch being processed. Every epoch the loss data will be printed to the console and the models weights and loss data will be saved to a timestamped folder inside of a directory named outputs that will be created automatically in the root directory.

Network Architecture

```

Layer (type:depth-idx) Output Shape Param #

FasterRCNN [100, 4] -- ├─GeneralizedRCNNTransform: 1-1 [16, 3, 1024, 800] -- ├─BackboneWithFPN: 1-2 [16, 256, 16, 13] -- │ └─IntermediateLayerGetter: 2-1 [16, 2048, 32, 25] -- │ │ └─Conv2d: 3-1 16, 64, 512, 400 │ │ └─FrozenBatchNorm2d: 3-2 [16, 64, 512, 400] -- │ │ └─ReLU: 3-3 [16, 64, 512, 400] -- │ │ └─MaxPool2d: 3-4 [16, 64, 256, 200] -- │ │ └─Sequential: 3-5 16, 256, 256, 200 │ │ └─Sequential: 3-6 [16, 512, 128, 100] 1,212,416 │ │ └─Sequential: 3-7 [16, 1024, 64, 50] 7,077,888 │ │ └─Sequential: 3-8 [16, 2048, 32, 25] 14,942,208 │ └─FeaturePyramidNetwork: 2-2 [16, 256, 16, 13] -- │ │ └─ModuleList: 3-15 -- (recursive) │ │ └─ModuleList: 3-16 -- (recursive) │ │ └─ModuleList: 3-15 -- (recursive) │ │ └─ModuleList: 3-16 -- (recursive) │ │ └─ModuleList: 3-15 -- (recursive) │ │ └─ModuleList: 3-16 -- (recursive) │ │ └─ModuleList: 3-15 -- (recursive) │ │ └─ModuleList: 3-16 -- (recursive) │ │ └─LastLevelMaxPool: 3-17 [16, 256, 256, 200] -- ├─RegionProposalNetwork: 1-3 [1000, 4] -- │ └─RPNHead: 2-3 [16, 3, 256, 200] -- │ │ └─Sequential: 3-18 [16, 256, 256, 200] 590,080 │ │ └─Conv2d: 3-19 [16, 3, 256, 200] 771 │ │ └─Conv2d: 3-20 [16, 12, 256, 200] 3,084 │ │ └─Sequential: 3-21 16, 256, 128, 100 │ │ └─Conv2d: 3-22 16, 3, 128, 100 │ │ └─Conv2d: 3-23 16, 12, 128, 100 │ │ └─Sequential: 3-24 16, 256, 64, 50 │ │ └─Conv2d: 3-25 16, 3, 64, 50 │ │ └─Conv2d: 3-26 16, 12, 64, 50 │ │ └─Sequential: 3-27 16, 256, 32, 25 │ │ └─Conv2d: 3-28 16, 3, 32, 25 │ │ └─Conv2d: 3-29 16, 12, 32, 25 │ │ └─Sequential: 3-30 16, 256, 16, 13 │ │ └─Conv2d: 3-31 16, 3, 16, 13 │ │ └─Conv2d: 3-32 16, 12, 16, 13 │ └─AnchorGenerator: 2-4 [204624, 4] -- ├─RoIHeads: 1-4 [100, 4] -- │ └─MultiScaleRoIAlign: 2-5 [16000, 256, 7, 7] -- │ └─TwoMLPHead: 2-6 [16000, 1024] -- │ │ └─Linear: 3-33 [16000, 1024] 12,846,080 │ │ └─Linear: 3-34 [16000, 1024] 1,049,600 │ └─FastRCNNPredictor: 2-7 [16000, 2] -- │ │ └─Linear: 3-35 [16000, 2] 2,050

│ │ └─Linear: 3-36 [16000, 8] 8,200

Total params: 41,299,161 Trainable params: 41,076,761 Non-trainable params: 222,400

Total mult-adds (Units.TERABYTES): 2.68

Input size (MB): 251.66 Forward/backward pass size (MB): 30311.83 Params size (MB): 165.20

Estimated Total Size (MB): 30728.68

```