Keras Computer Vision and NLP

Srikanth Prabhu
1 min readMar 28, 2024

--

Imagine classification with Keras is a deeper topic that needs to be discussed.

This can be used for also generating images. This is basically Stable Diffusion GenAI — image generation model. Stable diffusion is a strong open source text to image model, developed by StabilityAI.

Input = prompt

Output = amazing images.

How does it work? — There are a lot of networks which work ina. diffuser model. This will be used in a randomly generated diffuser model, cutting out all the noise.

There are several open source models available and Keras CV enjoys few of such models. This used NVIDIA F16ops which helps in using the graphics of the chip and making the model in place.

Tensorflow tooling allows everything be more smoother and everything at one place. Next is AppliedML with CV and NLP. We shall start with inference and fine-tune a pretrain large language model.

Inference — KerasCV offers a lot of pretrained models like imageNet, ClassificationNet. How should we make this to work — just by few lines of code, we can use a pretrained model of efficientNet.

Training an image classifier: We use a dataset and we split and shuffle a dataset. This has different sizes for every type of an image. We stack and build on top of EfficientNetV2.

Data Augumentation — Augmenting image data, Keras offers a wide layers of pretrained data augmentation pipelines. We preprocess the dataset by shuffling.
Pretrained layers like: RandAugment, CutMix, MixUp are used.

--

--

No responses yet