remove unnecessary stack in fast image processors and video processors

yonigozlan · yonigozlan · commit 5be88b50267d · 2025-12-09T20:36:52.000Z
diff --git a/src/transformers/models/beit/image_processing_beit_fast.py b/src/transformers/models/beit/image_processing_beit_fast.py
@@ -163,7 +163,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/bridgetower/image_processing_bridgetower_fast.py b/src/transformers/models/bridgetower/image_processing_bridgetower_fast.py
@@ -251,10 +251,8 @@ def _preprocess(
             processed_images, processed_masks = self.pad(
                 processed_images, return_mask=True, disable_grouping=disable_grouping
             )
-            processed_masks = torch.stack(processed_masks, dim=0) if return_tensors else processed_masks
             data["pixel_mask"] = processed_masks
 
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
         data["pixel_values"] = processed_images
 
         return BatchFeature(data=data, tensor_type=return_tensors)
diff --git a/src/transformers/models/cohere2_vision/image_processing_cohere2_vision_fast.py b/src/transformers/models/cohere2_vision/image_processing_cohere2_vision_fast.py
@@ -262,7 +262,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(
             data={"pixel_values": processed_images, "num_patches": num_patches}, tensor_type=return_tensors
diff --git a/src/transformers/models/convnext/image_processing_convnext_fast.py b/src/transformers/models/convnext/image_processing_convnext_fast.py
@@ -162,7 +162,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/deepseek_vl/image_processing_deepseek_vl_fast.py b/src/transformers/models/deepseek_vl/image_processing_deepseek_vl_fast.py
@@ -171,7 +171,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/deepseek_vl_hybrid/image_processing_deepseek_vl_hybrid_fast.py b/src/transformers/models/deepseek_vl_hybrid/image_processing_deepseek_vl_hybrid_fast.py
@@ -208,7 +208,6 @@ def _preprocess(
             high_res_processed_images_grouped[shape] = stacked_high_res_images
         high_res_processed_images = reorder_images(high_res_processed_images_grouped, grouped_high_res_images_index)
         high_res_processed_images = (
-            torch.stack(high_res_processed_images, dim=0) if return_tensors else high_res_processed_images
         )
 
         resized_images_grouped = {}
@@ -233,7 +232,6 @@ def _preprocess(
             )
             processed_images_grouped[shape] = stacked_images
         processed_images = reorder_images(processed_images_grouped, grouped_resized_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(
             data={"pixel_values": processed_images, "high_res_pixel_values": high_res_processed_images},
diff --git a/src/transformers/models/deepseek_vl_hybrid/modular_deepseek_vl_hybrid.py b/src/transformers/models/deepseek_vl_hybrid/modular_deepseek_vl_hybrid.py
@@ -889,7 +889,6 @@ def _preprocess(
             high_res_processed_images_grouped[shape] = stacked_high_res_images
         high_res_processed_images = reorder_images(high_res_processed_images_grouped, grouped_high_res_images_index)
         high_res_processed_images = (
-            torch.stack(high_res_processed_images, dim=0) if return_tensors else high_res_processed_images
         )
 
         resized_images_grouped = {}
@@ -914,7 +913,6 @@ def _preprocess(
             )
             processed_images_grouped[shape] = stacked_images
         processed_images = reorder_images(processed_images_grouped, grouped_resized_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(
             data={"pixel_values": processed_images, "high_res_pixel_values": high_res_processed_images},
diff --git a/src/transformers/models/depth_pro/image_processing_depth_pro_fast.py b/src/transformers/models/depth_pro/image_processing_depth_pro_fast.py
@@ -94,7 +94,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/dinov3_vit/image_processing_dinov3_vit_fast.py b/src/transformers/models/dinov3_vit/image_processing_dinov3_vit_fast.py
@@ -88,7 +88,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/donut/image_processing_donut_fast.py b/src/transformers/models/donut/image_processing_donut_fast.py
@@ -231,7 +231,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/dpt/image_processing_dpt_fast.py b/src/transformers/models/dpt/image_processing_dpt_fast.py
@@ -225,7 +225,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
         return BatchFeature(data={"pixel_values": processed_images})
 
     def post_process_semantic_segmentation(self, outputs, target_sizes: Optional[list[tuple]] = None):
diff --git a/src/transformers/models/dpt/modular_dpt.py b/src/transformers/models/dpt/modular_dpt.py
@@ -228,7 +228,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
         return BatchFeature(data={"pixel_values": processed_images})
 
     def post_process_depth_estimation(
diff --git a/src/transformers/models/efficientloftr/image_processing_efficientloftr_fast.py b/src/transformers/models/efficientloftr/image_processing_efficientloftr_fast.py
@@ -153,7 +153,6 @@ def _preprocess(
         stacked_pairs = [torch.stack(pair, dim=0) for pair in image_pairs]
 
         # Return in same format as slow processor
-        image_pairs = torch.stack(stacked_pairs, dim=0) if return_tensors else stacked_pairs
 
         return BatchFeature(data={"pixel_values": image_pairs})
 
diff --git a/src/transformers/models/efficientnet/image_processing_efficientnet_fast.py b/src/transformers/models/efficientnet/image_processing_efficientnet_fast.py
@@ -178,7 +178,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/eomt/image_processing_eomt_fast.py b/src/transformers/models/eomt/image_processing_eomt_fast.py
@@ -276,7 +276,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
         images = reorder_images(processed_images_grouped, grouped_images_index)
 
-        processed_images = torch.stack(images, dim=0) if return_tensors else images
 
         return processed_images, patch_offsets
 
diff --git a/src/transformers/models/flava/image_processing_flava_fast.py b/src/transformers/models/flava/image_processing_flava_fast.py
@@ -306,7 +306,6 @@ def _preprocess_image(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return processed_images
 
@@ -397,7 +396,6 @@ def _preprocess(
                 mask_group_max_aspect_ratio=mask_group_max_aspect_ratio,
             )
             masks = [mask_generator() for _ in range(len(images))]
-            masks = torch.stack(masks, dim=0) if return_tensors else masks
             data["bool_masked_pos"] = masks
 
         return BatchFeature(data=data, tensor_type=return_tensors)
diff --git a/src/transformers/models/glpn/image_processing_glpn_fast.py b/src/transformers/models/glpn/image_processing_glpn_fast.py
@@ -107,7 +107,6 @@ def _preprocess(
             processed_groups[shape] = stacked_images
 
         processed_images = reorder_images(processed_groups, grouped_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
     def post_process_depth_estimation(self, outputs, target_sizes=None):
diff --git a/src/transformers/models/got_ocr2/image_processing_got_ocr2_fast.py b/src/transformers/models/got_ocr2/image_processing_got_ocr2_fast.py
@@ -189,7 +189,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(
             data={"pixel_values": processed_images, "num_patches": num_patches}, tensor_type=return_tensors
diff --git a/src/transformers/models/imagegpt/image_processing_imagegpt_fast.py b/src/transformers/models/imagegpt/image_processing_imagegpt_fast.py
@@ -165,11 +165,9 @@ def _preprocess(
             input_ids = reorder_images(input_ids_grouped, grouped_images_index)
 
             return BatchFeature(
-                data={"input_ids": torch.stack(input_ids, dim=0) if return_tensors else input_ids},
                 tensor_type=return_tensors,
             )
 
-        pixel_values = torch.stack(pixel_values, dim=0) if return_tensors else pixel_values
         return BatchFeature(data={"pixel_values": pixel_values}, tensor_type=return_tensors)
 
     def to_dict(self):
diff --git a/src/transformers/models/instructblipvideo/video_processing_instructblipvideo.py b/src/transformers/models/instructblipvideo/video_processing_instructblipvideo.py
@@ -84,7 +84,6 @@ def _preprocess(
             processed_videos_grouped[shape] = stacked_videos
 
         processed_videos = reorder_videos(processed_videos_grouped, grouped_videos_index)
-        processed_videos = torch.stack(processed_videos, dim=0) if return_tensors else processed_videos
 
         return BatchFeature(data={"pixel_values": processed_videos}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/internvl/video_processing_internvl.py b/src/transformers/models/internvl/video_processing_internvl.py
@@ -140,7 +140,6 @@ def _preprocess(
             processed_videos_grouped[shape] = stacked_videos
 
         processed_videos = reorder_videos(processed_videos_grouped, grouped_videos_index)
-        processed_videos = torch.stack(processed_videos, dim=0) if return_tensors else processed_videos
 
         return BatchFeature(data={"pixel_values_videos": processed_videos}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/janus/image_processing_janus_fast.py b/src/transformers/models/janus/image_processing_janus_fast.py
@@ -180,7 +180,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/kosmos2_5/image_processing_kosmos2_5_fast.py b/src/transformers/models/kosmos2_5/image_processing_kosmos2_5_fast.py
@@ -264,8 +264,6 @@ def _preprocess(
 
         encoded_outputs = BatchFeature(
             data={
-                "flattened_patches": torch.stack(flattened_patches, dim=0) if return_tensors else flattened_patches,
-                "attention_mask": torch.stack(attention_masks, dim=0) if return_tensors else attention_masks,
                 "width": width,
                 "height": height,
                 "rows": rows,
diff --git a/src/transformers/models/layoutlmv2/image_processing_layoutlmv2_fast.py b/src/transformers/models/layoutlmv2/image_processing_layoutlmv2_fast.py
@@ -101,7 +101,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         data = BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/layoutlmv3/image_processing_layoutlmv3_fast.py b/src/transformers/models/layoutlmv3/image_processing_layoutlmv3_fast.py
@@ -115,7 +115,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         data = BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/lightglue/image_processing_lightglue_fast.py b/src/transformers/models/lightglue/image_processing_lightglue_fast.py
@@ -174,7 +174,6 @@ def _preprocess(
         stacked_pairs = [torch.stack(pair, dim=0) for pair in image_pairs]
 
         # Return in same format as slow processor
-        image_pairs = torch.stack(stacked_pairs, dim=0) if return_tensors else stacked_pairs
 
         return BatchFeature(data={"pixel_values": image_pairs})
 
diff --git a/src/transformers/models/llama4/image_processing_llama4_fast.py b/src/transformers/models/llama4/image_processing_llama4_fast.py
@@ -422,7 +422,6 @@ def _preprocess(
         aspect_ratios_list = reorder_images(grouped_aspect_ratios, grouped_images_index)
 
         processed_images = torch.cat(processed_images, dim=0) if return_tensors else processed_images
-        aspect_ratios = torch.stack(aspect_ratios_list, dim=0) if return_tensors else aspect_ratios_list
         return BatchFeature(
             data={"pixel_values": processed_images, "aspect_ratios": aspect_ratios}, tensor_type=return_tensors
         )
diff --git a/src/transformers/models/llava/image_processing_llava_fast.py b/src/transformers/models/llava/image_processing_llava_fast.py
@@ -149,7 +149,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/llava_next/image_processing_llava_next_fast.py b/src/transformers/models/llava_next/image_processing_llava_next_fast.py
@@ -260,7 +260,6 @@ def _preprocess(
 
         if do_pad:
             processed_images = self._pad_for_batching(processed_images)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
         return BatchFeature(
             data={"pixel_values": processed_images, "image_sizes": image_sizes}, tensor_type=return_tensors
         )
diff --git a/src/transformers/models/llava_onevision/image_processing_llava_onevision_fast.py b/src/transformers/models/llava_onevision/image_processing_llava_onevision_fast.py
@@ -279,7 +279,6 @@ def _preprocess(
 
         if do_pad:
             processed_images = self._pad_for_batching(processed_images)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
         return BatchFeature(
             data={"pixel_values": processed_images, "image_sizes": image_sizes, "batch_num_images": batch_num_images},
             tensor_type=return_tensors,
diff --git a/src/transformers/models/llava_onevision/modular_llava_onevision.py b/src/transformers/models/llava_onevision/modular_llava_onevision.py
@@ -211,7 +211,6 @@ def _preprocess(
 
         if do_pad:
             processed_images = self._pad_for_batching(processed_images)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
         return BatchFeature(
             data={"pixel_values": processed_images, "image_sizes": image_sizes, "batch_num_images": batch_num_images},
             tensor_type=return_tensors,
diff --git a/src/transformers/models/mask2former/image_processing_mask2former_fast.py b/src/transformers/models/mask2former/image_processing_mask2former_fast.py
@@ -388,8 +388,6 @@ def _preprocess(
         processed_pixel_masks = reorder_images(processed_pixel_masks_grouped, grouped_images_index)
         encoded_inputs = BatchFeature(
             data={
-                "pixel_values": torch.stack(processed_images, dim=0) if return_tensors else processed_images,
-                "pixel_mask": torch.stack(processed_pixel_masks, dim=0) if return_tensors else processed_pixel_masks,
             },
             tensor_type=return_tensors,
         )
diff --git a/src/transformers/models/maskformer/image_processing_maskformer_fast.py b/src/transformers/models/maskformer/image_processing_maskformer_fast.py
@@ -392,8 +392,6 @@ def _preprocess(
         processed_pixel_masks = reorder_images(processed_pixel_masks_grouped, grouped_images_index)
         encoded_inputs = BatchFeature(
             data={
-                "pixel_values": torch.stack(processed_images, dim=0) if return_tensors else processed_images,
-                "pixel_mask": torch.stack(processed_pixel_masks, dim=0) if return_tensors else processed_pixel_masks,
             },
             tensor_type=return_tensors,
         )
diff --git a/src/transformers/models/mobilenet_v2/image_processing_mobilenet_v2_fast.py b/src/transformers/models/mobilenet_v2/image_processing_mobilenet_v2_fast.py
@@ -180,7 +180,6 @@ def _preprocess(
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
 
         # Stack all processed images if return_tensors is specified
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/mobilevit/image_processing_mobilevit_fast.py b/src/transformers/models/mobilevit/image_processing_mobilevit_fast.py
@@ -182,7 +182,6 @@ def _preprocess(
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
 
         # Stack all processed images if return_tensors is specified
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/nougat/image_processing_nougat_fast.py b/src/transformers/models/nougat/image_processing_nougat_fast.py
@@ -290,7 +290,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/ovis2/image_processing_ovis2_fast.py b/src/transformers/models/ovis2/image_processing_ovis2_fast.py
@@ -213,7 +213,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
         return BatchFeature(data={"pixel_values": processed_images, "grids": grids}, tensor_type=return_tensors)
 
 
diff --git a/src/transformers/models/owlv2/image_processing_owlv2_fast.py b/src/transformers/models/owlv2/image_processing_owlv2_fast.py
@@ -336,7 +336,6 @@ def _preprocess(
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
 
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/owlv2/modular_owlv2.py b/src/transformers/models/owlv2/modular_owlv2.py
@@ -205,7 +205,6 @@ def _preprocess(
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
 
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/perceiver/image_processing_perceiver_fast.py b/src/transformers/models/perceiver/image_processing_perceiver_fast.py
@@ -113,7 +113,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/perception_lm/image_processing_perception_lm_fast.py b/src/transformers/models/perception_lm/image_processing_perception_lm_fast.py
@@ -307,7 +307,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
         processed_images = [p[None] if p.ndim == 3 else p for p in processed_images]  # add tiles dimension if needed
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
 
diff --git a/src/transformers/models/poolformer/image_processing_poolformer_fast.py b/src/transformers/models/poolformer/image_processing_poolformer_fast.py
@@ -231,7 +231,6 @@ def _preprocess(
             processed_images_grouped[shape] = stacked_images
 
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/sam/image_processing_sam_fast.py b/src/transformers/models/sam/image_processing_sam_fast.py
@@ -267,7 +267,6 @@ def _preprocess(
         if do_pad:
             processed_images = self.pad(processed_images, pad_size=pad_size, disable_grouping=disable_grouping)
 
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
         return BatchFeature(
             data={"pixel_values": processed_images, "reshaped_input_sizes": reshaped_input_sizes},
             tensor_type=return_tensors,
diff --git a/src/transformers/models/segformer/image_processing_segformer_fast.py b/src/transformers/models/segformer/image_processing_segformer_fast.py
@@ -168,7 +168,6 @@ def _preprocess(
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
 
         # Stack images into a single tensor if return_tensors is set
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/segformer/modular_segformer.py b/src/transformers/models/segformer/modular_segformer.py
@@ -140,7 +140,6 @@ def _preprocess(
         processed_images = reorder_images(processed_images_grouped, grouped_images_index)
 
         # Stack images into a single tensor if return_tensors is set
-        processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images
 
         return BatchFeature(data={"pixel_values": processed_images}, tensor_type=return_tensors)
 
diff --git a/src/transformers/models/smolvlm/video_processing_smolvlm.py b/src/transformers/models/smolvlm/video_processing_smolvlm.py
diff --git a/src/transformers/models/superglue/image_processing_superglue_fast.py b/src/transformers/models/superglue/image_processing_superglue_fast.py
diff --git a/src/transformers/models/superpoint/image_processing_superpoint_fast.py b/src/transformers/models/superpoint/image_processing_superpoint_fast.py
diff --git a/src/transformers/models/swin2sr/image_processing_swin2sr_fast.py b/src/transformers/models/swin2sr/image_processing_swin2sr_fast.py
diff --git a/src/transformers/models/textnet/image_processing_textnet_fast.py b/src/transformers/models/textnet/image_processing_textnet_fast.py
diff --git a/src/transformers/models/vitmatte/image_processing_vitmatte_fast.py b/src/transformers/models/vitmatte/image_processing_vitmatte_fast.py
diff --git a/src/transformers/models/vitpose/image_processing_vitpose_fast.py b/src/transformers/models/vitpose/image_processing_vitpose_fast.py
diff --git a/src/transformers/models/zoedepth/image_processing_zoedepth_fast.py b/src/transformers/models/zoedepth/image_processing_zoedepth_fast.py
diff --git a/src/transformers/video_processing_utils.py b/src/transformers/video_processing_utils.py

Original file line number	Diff line number	Diff line change
`@@ -251,10 +251,8 @@ def _preprocess(`
`251`	`251`	`processed_images, processed_masks = self.pad(`
`252`	`252`	`processed_images, return_mask=True, disable_grouping=disable_grouping`
`253`	`253`	`)`
`254`		`- processed_masks = torch.stack(processed_masks, dim=0) if return_tensors else processed_masks`
`255`	`254`	`data["pixel_mask"] = processed_masks`
`256`	`255`
`257`		`- processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images`
`258`	`256`	`data["pixel_values"] = processed_images`
`259`	`257`
`260`	`258`	`return BatchFeature(data=data, tensor_type=return_tensors)`
Original file line number	Diff line number	Diff line change
`@@ -208,7 +208,6 @@ def _preprocess(`
`208`	`208`	`high_res_processed_images_grouped[shape] = stacked_high_res_images`
`209`	`209`	`high_res_processed_images = reorder_images(high_res_processed_images_grouped, grouped_high_res_images_index)`
`210`	`210`	`high_res_processed_images = (`
`211`		`- torch.stack(high_res_processed_images, dim=0) if return_tensors else high_res_processed_images`
`212`	`211`	`)`
`213`	`212`
`214`	`213`	`resized_images_grouped = {}`
`@@ -233,7 +232,6 @@ def _preprocess(`
`233`	`232`	`)`
`234`	`233`	`processed_images_grouped[shape] = stacked_images`
`235`	`234`	`processed_images = reorder_images(processed_images_grouped, grouped_resized_images_index)`
`236`		`- processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images`
`237`	`235`
`238`	`236`	`return BatchFeature(`
`239`	`237`	`data={"pixel_values": processed_images, "high_res_pixel_values": high_res_processed_images},`
Original file line number	Diff line number	Diff line change
`@@ -889,7 +889,6 @@ def _preprocess(`
`889`	`889`	`high_res_processed_images_grouped[shape] = stacked_high_res_images`
`890`	`890`	`high_res_processed_images = reorder_images(high_res_processed_images_grouped, grouped_high_res_images_index)`
`891`	`891`	`high_res_processed_images = (`
`892`		`- torch.stack(high_res_processed_images, dim=0) if return_tensors else high_res_processed_images`
`893`	`892`	`)`
`894`	`893`
`895`	`894`	`resized_images_grouped = {}`
`@@ -914,7 +913,6 @@ def _preprocess(`
`914`	`913`	`)`
`915`	`914`	`processed_images_grouped[shape] = stacked_images`
`916`	`915`	`processed_images = reorder_images(processed_images_grouped, grouped_resized_images_index)`
`917`		`- processed_images = torch.stack(processed_images, dim=0) if return_tensors else processed_images`
`918`	`916`
`919`	`917`	`return BatchFeature(`
`920`	`918`	`data={"pixel_values": processed_images, "high_res_pixel_values": high_res_processed_images},`