I used the Vertex AI Model Garden notebook to upload a DeepLabv3+ model to the model registry. As I realized, my input data in JSONL format should include each image in base 64 encoding on each line. Upon trying, it seems I cannot provide the input in any other form in JSONL. The ideal scenario would be providing the input as filenames. More importantly, I need to figure out how to associate segmentation results with their respective image URIs when sending multiple images through a JSONL file. Is there a way to send filenames in the JSONL file instead of encoded images? Alternatively, is there a specific JSON tag or field that I can include in each JSON object to specify the image name or identifier?
I am aware that I can use a file list as the input, but is that the only solution? If yes, how can I send a file list in a request to the endpoint?
LinkedIn
Twitter
