When I try to use DPT_SwinV2_L_384 backbone I am getting shape mismatch errors. The part of the code where the Midas model is build and the weights are loaded executes without any problem. However, when ZoeDepth model is build using the Midas core model, during the loading part of the weights, shape mismatch between the pre-trained weights and the model occurs.
I was wondering whether the repository has been tested with different backbones of the Midas models or it has only been tested using DPT_BEiT_L_384 backbone?
When I try to use DPT_SwinV2_L_384 backbone I am getting shape mismatch errors. The part of the code where the Midas model is build and the weights are loaded executes without any problem. However, when ZoeDepth model is build using the Midas core model, during the loading part of the weights, shape mismatch between the pre-trained weights and the model occurs.
I was wondering whether the repository has been tested with different backbones of the Midas models or it has only been tested using DPT_BEiT_L_384 backbone?