Different applications of target re-identification (Re-ID), such as threat identification and access control, have been studied extensively, but vehicle Re-ID presents more complex challenges. The differences in appearance of the same vehicle from multiple cameras and angles makes the task of recognizing a vehicle of interest difficult. This model, however, uses powerful techniques that utilize these varying perspectives to its advantage. This model re-identifies vehicles by comparing the vehicle of interest (“query image”) to a database of vehicles (“gallery” images) and returns the probability of a match. The set of gallery images are also provided, making the model versatile in its matching capabilities which is a valued quality for Re-ID models.
See the model in action with a Modzy MLOps platform demo or start a trial
87.7% Accuracy
87.6% F1 Score
87.9% Precision
87.3% Recall
This model was validated on similar and dissimilar pairs of vehicles from the VERI dataset, a large-scale benchmark dataset specifically curated for vehicle Re-ID. At a confidence threshold of 0.4, the model achieves an accuracy of 0.8766, precision of 0.8792, recall of 0.8732, and an F1 score of 0.8762.
The fraction of correct predictions made by the classifier. This metric is calculated by dividing the number of correct predictions by the total number of predictions.
Is the harmonic mean of the precision and recall, with best value of 1. It measures the balance between the two metrics.
A higher precision score indicates that on average the majority of labels predicted by the model for different classes are accurate.
A higher recall score indicates that the model finds and predicts correct labels for the majority of the classes it is supposed to find.
This model uses a Pyramid Granularity Attentive Model (PGAM) as its architecture. Specifically, it implements a multi-scale pyramid design, extracting features on both the coarse scale and fine grain levels. This method ultimately leads to the extraction of both global and local features in multiple scales, improving model Re-ID performance with respect to other methods that directly separate global and local features.
This model was trained and validated on the VERI dataset, which contains a diverse set of over 50,000 images of 776 vehicles, captured by 20 cameras, which cover a 1.0 km^2 area in 24 hours. During training, the following strategies were used to improve performance: random erasing augmentation, Batch-Normalization Neck design, and center loss.
This model was validated on similar and dissimilar pairs of vehicles from the VERI dataset with a confidence threshold of 0.4.
The input(s) to this model must adhere to the following specifications:
The zip input file must contain both query and gallery images, where a query image is defined as the image of interest to be compared to a set of gallery images from an existing database. The model will only process images within the zip file that contain “query” or “gallery” in their respective filepaths. For example, the filepaths “/data/query.jpg”, “/query/image1.jpg”, “/data/test/gallery/img.png” would all be processed, but the filepath “/data/q/image.jpg” would not.
This model will output the following:
The output JSON file contains a list of similarity scores between the query input(s) and gallery input(s).
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
