ViTObjectDetector constructor

ViTObjectDetector({
  1. required int imageSize,
  2. required int patchSize,
  3. required int numChannels,
  4. required int embedSize,
  5. required int numLayers,
  6. required int numHeads,
  7. required int numClasses,
  8. required int numQueries,
})

Implementation

ViTObjectDetector({
  required int imageSize,
  required int patchSize,
  required int numChannels,
  required this.embedSize,
  required int numLayers,
  required int numHeads,
  required this.numClasses, // Store this to reshape correctly
  required this.numQueries,
}) : backbone = ViTBackbone(
       imageSize: imageSize,
       patchSize: patchSize,
       numChannels: numChannels,
       embedSize: embedSize,
       numLayers: numLayers,
       numHeads: numHeads,
     ) {
  classHead = Layer(embedSize, numQueries * (numClasses + 1), useGelu: false);
  boxHead = Layer(embedSize, numQueries * 4, useGelu: false);
}