This model classifies each object in an image into one of 1,000 general object classes, which are organized in a hierarchy, available in the ImageNet dataset. The model can also defend against possible adversarial attacks. This model accepts JPEG images of any size. The model outputs the top five predictions that have the highest confidence, along with the explainability object for the respective predictions, in a JSON file that also includes object hierarchy information.
This model can be used to classify a large set of images into different hierarchical categories based on object classes presented in the images.
See the model in action with a Modzy MLOps platform demo or start a trial
Adversarial Defense: This model has a built-in adversarial defense feature. Learn more about Modzy Adversarial Defense.
Explainable: This model has a built-in explainability feature. What is model explainability?.
67% Top 1 Accuracy
86% Top 5 Accuracy
68% Top 1 Robustness Accuracy
86% Top 5 Robustness Accuracy
This model was trained on the ImageNet dataset. The model was tested on the ImageNet test dataset of 50,000 images. The model obtains the state-of-the-art top 1 accuracy of 0.67 and top 5 accuracy of 0.86 on the test dataset, and top 1 robustness accuracy of 0.68 and top 5 robustness accuracy of 0.86 on the synthetically generated adversarial dataset. Some of this model’s strengths include its capability in detecting and classifying object classes for a large range of images of different sizes and pixel distributions with the ability to find the top 5 appropriate object classes presented in the image. For fast inference time, this model works best on a GPU.
Top 1 Accuracy: The ratio of the number of correct predictions of the top 1 predicted class to the total number of input samples.
Top 5 Accuracy: The fraction of the top 5 predictions made by the classifier.
Top 1 Robustness Accuracy: The accuracy of the classifier on synthetically generated adversarial dataset.
Top 5 Robustness Accuracy: The fraction of the top 5 accuracies made by the classifier on synthetically generated adversarial test dataset.
This model utilizes a residual neural network called ResNet50, which was first developed and published by Microsoft Research. ResNet50 uses residual learning, which leads to networks that are very deep and able to learn a large set of features from the training dataset. The features learned by this deep learning model are low-level, mid-level and high-level features presented at the end of each sub-block inside the network. These features are residual features that can be simply understood as subtraction of features learned by the previous layer. The features learned by this model are not only representative of the object classes in images but are also robust against adversarial attacks. This model was developed using the TensorFlow deep learning framework.
This model was trained on the ImageNet dataset, a large imagery dataset of 1.2 million images and 1,000 different object classes. The model was trained using our Lyapunov based robust solution. The training took 51 days on 1 DGX-1 GPU. The model was trained using stochastic gradient descent with a learning rate of 0.01 and momentum of 0.9 for 100 epochs on 1.2 million training images.
The performance of the model was tested on a validation dataset of 50,000 images and obtained the state-of-the-art top 1 accuracy of 0.67 and top 5 accuracy of 0.86. The model also obtained the state-of-the-art top 1 robustness accuracy of 0.68 and top 5 robustness accuracy of 0.86 on the synthetically generated adversarial dataset of 50,000 adversarial images generated by a range of different adversarial attacks.
The input(s) to this model must adhere to the following specifications:
This model will output the following:
See how quickly you can deploy and run models, connect to pipelines, autoscale resources, and integrate into workflows with Modzy MLOps platform
d o n o t fill t h i s . f i e l d d o n o t fill t h i s . f i e l d
