Saliency map uniformly red in blackbox testing

[cansik]

I tried to use OpenVINO-XAI on my quantized YoloX object detection model, which has not been trained with the training_extensions but is an implementation from super-gradients. I was able to make it work with OpenVINO-XAI, and I am getting correct bounding boxes, scores, and class labels. However, the saliency map appears to be uniformly red (0, 0, 128) after colorization by the explainer (using black-box testing).

self.xai_explainer = xai.Explainer(model=self.engine.model.path,
                                               task=xai.Task.DETECTION,
                                               explain_mode=ExplainMode.BLACKBOX,
                                               preprocess_fn=self._pre_process_fn,
                                               postprocess_fn=self._post_process_fn)

explanation: Explanation = self.xai_explainer.explain(data=image,
                                                      targets=-1,
                                                      preset=Preset.SPEED
                                                      )

What could be causing this issue?
And what would you need to investigate this or how can I further investigate the issue?

It is currently not clear to me, what I could change to get further insights into the explanation process. I was not able to find a lot of information about the xai.Task.DETECTION mode, most of the notebooks are targeted for the classification task.