Parse raw model output to structured JSON object #19
Conversation
There was a problem hiding this comment.
Good work so far with parsing model output to keys! I would say as a temporary measure for the demo, if it is possible, please add the full generated text to the repsonse since after parsing to keys a lot of the transcribed text ends up missing (likely as mentioned due to model performance)
| @@ -16,6 +18,7 @@ | |||
|
|
|||
| @app.route("/api/transcribe", methods=["POST"]) | |||
| def transcribe(): | |||
| print("START OF ENDPOINT") | |||
| @@ -0,0 +1,18 @@ | |||
| { | |||
There was a problem hiding this comment.
For now this is okay as this model's accuracy is low and we will also need to consider other database templates, but we will need a more comprehensive list of keys for all subheadings in log template
| return jsonify({"transcription": generated_text}) | ||
| keys = load_keys("keys.json") | ||
| json_result = parse_florence_output(generated_text, keys) | ||
| return json_result |
There was a problem hiding this comment.
Due to low model accuracy, could you please also include the full generated_text in the API's repsonse for demo purposes? I could maybe suggest return a jsonify or json.dumps that combines json_result with generated_text
Note: