Notes in week3

To Subscribe, use this Key

Status	Last Update	Fields
Published	04/07/2024	Types of Image segmentation: semantic segmentation: {{c1::assigning a class label to each individual pixel.}}instance segmentation: {{c2::assigns…
Published	04/07/2024	Object detection network architectureFor object detection we use a {{c1::Fully Convolutional Network to classify each pixel}}. A Limitation: {{c2…
Published	04/07/2024	Atrous convolutional kernel why do we use an atrous kernel ? {{c2::to increase the receptive field without adding more parameters}}how does …
Published	04/07/2024	Encoder & Decoderswhat is the purpose of encoder?{{c1::the encoder shrinks the image dimensions, to increase the receptive field.}}what is the pur…
Published	04/07/2024	Transposed convolutionmethod of increasing the input dimensions (upsampling). by ? {{c1::multiplying a single value with all the kernel values, a…
Published	04/07/2024	Max UnpoolingRevert the maxpooling operation (thus upsample).Remember the {{c1::max value index}} to place back the max value to itst original in…
Published	04/07/2024	Billinear upsamplingSpread input values and use {{c1::linear interpolations i.e. generate values that are inbetween the input values}}.
Published	04/07/2024	Upsampling through Pixel Shuffle{{c1::Use values from multiple feature maps to increase resolution.}}
Published	04/07/2024	U-NetSemantic segmenation model, for brain scansuses encoder & decoders (U-shape)Why does bilinear = False require more training parameters?{{c1::…
Published	04/07/2024	AlexNetwhy is the input 227x227x3? {{c1::RGB images}}Given kernel = 11x11, feature maps = 96, why has conv1 34944 parameters? {{c2:: we…
Published	04/07/2024	VGG networkmost often used modelDeep convolutional networkslow traininglarge model (500 mb)gradient problem
Published	04/07/2024	ResNet Dealt with the gradient problem and degradatioin problem by using {{c1::residual blocks, by using a skip connection (identity function).}}…
Published	04/07/2024	Receptive fieldterm to indicate: {{c1::how many pixels does a layer see}}given two 3x3 convolutions what is the receptive field? {{c2::RF = 5 bec…
Published	04/07/2024	8e258722b84a44538ce904cde702d787-ao-1
Published	04/07/2024	20d55d364ccc4bebae8709c754dd6e91-ao-1
Published	04/07/2024	20d55d364ccc4bebae8709c754dd6e91-ao-2
Published	04/07/2024	20d55d364ccc4bebae8709c754dd6e91-ao-3
Published	04/07/2024	20d55d364ccc4bebae8709c754dd6e91-ao-4
Published	04/07/2024	20d55d364ccc4bebae8709c754dd6e91-ao-5
Published	04/07/2024	Sliding window for object detection + classificationwhy was this approach inefficient? {{c1:: as the sliding window had to be repeated many times…
Published	04/07/2024	Regional CNN (R-CNN / RPN)why is it a two-stage detector {{c1::becuase it uses a model for making proposal and a model for classification}}what is the…
Published	04/07/2024	Common R-CNNsHow did Faster R-CNN improve over Fast R-CNN? {{c1:: It uses a single CNN for generating Region of Interests (ROIs) and features. Wh…
Published	04/07/2024	You Only Look Once (YOLO) Why is YOLO as single-stage model ? {{c1::because it uses a single pipeline for detection and classification.}}met…
Published	04/07/2024	How to keep good predictions?{{c1::use a background class}}{{c1::threshold the confidence}}
Published	04/07/2024	Non-maximum Suppression (NMS) & Intersection over Union (IoU)NMS uses IoU to:{{c1::filter out bounding boxes that exceed the IoU overlap threshold…
Published	04/07/2024	Accuracy assesment of Object DetectorIoU: {{c1::compare detection bbox with ground truth bbox}}Average Precission (AP): {{c2::The area under…
Published	04/07/2024	Hungarian algorithmuses IoU to build an {{c1::adjecency matrix of predicted bboxes and ground truth bboxes}}to select the {{c2::optimal bbox…
Published	04/07/2024	Kaas = Kaas {{c1::Gouda}} {{c1::Maasdammer}}
Published	04/07/2024	Translation InvarianceWhat is the meaning of Translation? {{c1::Translation is displacement in space}}What is meant by translation invariance? {{…
Published	04/07/2024	0f41c720b9c845209af681ab8b51e678-oa-1
Published	04/07/2024	0f41c720b9c845209af681ab8b51e678-oa-2
Published	04/07/2024	0f41c720b9c845209af681ab8b51e678-oa-3
Published	04/07/2024	0f41c720b9c845209af681ab8b51e678-oa-4
Published	04/07/2024	0f41c720b9c845209af681ab8b51e678-oa-5
Published	04/07/2024	dcebc17267264b48b17aff4b04b5b90c-oa-1
Published	04/07/2024	dcebc17267264b48b17aff4b04b5b90c-oa-2
Published	04/07/2024	dcebc17267264b48b17aff4b04b5b90c-oa-3
Published	04/07/2024	dcebc17267264b48b17aff4b04b5b90c-oa-4
Published	04/07/2024	dcebc17267264b48b17aff4b04b5b90c-oa-5
Published	04/07/2024	dcebc17267264b48b17aff4b04b5b90c-oa-6
Published	04/07/2024	Anchor boxes
Status	Last Update	Fields