Add Torch Inferencer and Update Openvino and Gradio Inferencers. #453
Conversation
| @@ -141,9 +149,30 @@ def post_process( | |||
| anomaly_map = predictions.squeeze() | |||
| pred_score = anomaly_map.reshape(-1).max() | |||
|
|
|||
| # Common practice in anomaly detection is to assign anomalous | |||
| # label to the prediction if the prediction score is greater | |||
| # than the image threshold. | |||
There was a problem hiding this comment.
Should we warn users if thresholds are not present in meta_data?
There was a problem hiding this comment.
This would bring up another discussion. Do you think we should check if they are in meta_data?
For example,
if "image_threshold" in meta_data:
pred_label = pred_score >= meta_data["image_threshold"]
We cannot calculate the pred-label if we don't have this. The question is whether we should have this if statement here, and for the rest of the computation?
There was a problem hiding this comment.
I think the default threshold is 0.5 so if we don't have image_threshold in meta_data then we can throw a warning and reduce the if statement to pred_label = pred_score >= meta_data.get("image_threshold", 0.5)
There was a problem hiding this comment.
As long as we throw a warning it should be fine. Because the users need to be made aware that there is a problem with the metadata file.
| pred_mask = cv2.resize(pred_mask, (image_width, image_height)) | ||
|
|
||
| return { | ||
| "anomaly_map": anomaly_map, |
There was a problem hiding this comment.
Would this work for task type of classification?
There was a problem hiding this comment.
Yeah, it returns none, which ImageResults handles ok.
Description
Add Torch Inferencer.
Return all the predictions from
inferencer.predict(image). This way, the user will be able to access to any info they need.Fixes # 294
Changes
Checklist