Core demands
Car buyers and car enthusiasts have a very strong demand for car quotes and car information. Pacific Auto Network has hundreds of first-line senior professional editors to provide the latest information. At the same time, it provides the latest quotations in nearly 400 regions of the country, and updates more than 300 quotations every day. The model library covers more than 180 automobile brands, more than 25000 models of detailed data, and more than 3 million model pictures. There is a large amount of information, and there is a certain retrieval boundary through text search. Therefore, we hope to increase the search method of image recognition, increase the content of retrieval results, and indirectly improve the transformation.
Solution
In order to ensure the comprehensiveness and accuracy of the model recognition results, the Pacific Automotive Network information client uses Baidu's image recognition - model recognition technology and Baidu EasyDL customized new model recognition model. After uploading the picture, the user calls the vehicle identification and EasyDL customized vehicle identification interfaces at the same time. According to the vehicle identification and EasyDLtop1 classification results, the relevant vehicle models are displayed in the order of confidence from high to low, so that the user can choose to know the details according to the returned results.
The specific process is as follows:
Step 1: For the pictures uploaded by users, call Baidu model identification and EasyDL customized model identification interfaces at the same time.
Step 2: Set the priority list according to the EasyDL recognition list, and the confidence level of EasyDL shall prevail if the vehicle model recognition interface and EasyDL interface both hit.
Step 3: Rank the confidence from high to low according to the top5 results returned from the model recognition and the top1 model results from EasyDL.
Step 4: According to the model name returned from the image recognition interface, match the cover image and dealer quotation in the self owned library for front-end display.
An example of the front end photo identification process is as follows:
