How to remove reflections from Images?

I am trying to remove reflections from a Car. The problem I face is that another damage detection model detects the reflections as dents/scratches on the cars. Hence I have been working on a solution to try to reduce the amount of reflections on the car. I have tried these solution so far:

1. Obstruction Removal (https://github.com/alex04072000/ObstructionRemoval)This method takes multiple frames into account and tries to separate the reflection layer from the foreground layer. Running these on car images I found the results disappointing. The reflections were not removed. The only change I saw was a reduction in white balance. There was a certain reduction of brightness in some areas, but for the problem we face, this method does not help at all. Also this method requires multiple frames.

2. GCNET(https://github.com/ryo-abiko/GCNet) This method uses a GAN to remove reflections from single images. This method provided similar results to Obstruction removal. It reduced brightness in some areas. I feel this technique is more catered to removing white light reflections from the glass. I tried passing in only the car window images to this network. The reflections in the glass were still present. A few highlighted areas had reduced brightness.

3. ERRNet (https://github.com/Vandermode/ERRNet) This model was trained on single image pairs with reflective images and GT images. Running some samples from our dataset provided results that were not promising. Most of the reflections were still present.

I have tried all the top possible methods from paperswithcode. None have been giving satisfactory result. Is this outcome possible through any other way?

Example Images: