← Computer Vision · Repository
The image-classification area contains architecture-first TensorFlow projects.
Most runnable variants pair config.py, dataset.py, model.py, and
train_and_test.py; configuration selects supported local datasets and
training/runtime options. The pages below group depth and scale variants so
the documentation follows the implementation families rather than duplicating
the same workflow.
| Track | Families | Documentation |
|---|---|---|
| Small-network and efficiency-focused | LeNet-5, ZFNet, SqueezeNet, XNOR-Net, MobileNet, ShuffleNet, knowledge distillation, deep compression, FractalNet, MLP-Mixer, PolyNet, Xception | Small networks |
| Large convolutional and attention models | AlexNet, VGG, Network in Network, Inception, ResNet, HighwayNet, DenseNet, residual attention, SENet, ResNeXt, CapsuleNet, Vision Transformer | Large networks |
| Robustness | Adversarial saliency maps, black-box attacks, FGSM, iterative least-likely method | Robustness |
The family implementations construct architectures from TensorFlow/Keras
layers instead of tf.keras.applications. Input pipelines, data roots,
augmentation, optimizer choices, and output directories are controlled per
folder; not every variant has an identical set of datasets or a retained
experiment report. Follow the selected folder's scripts and configuration.