Removed 'slices' from EncodedImage

[popovaan]

ov::Tensor slices can be removed from EncodedImage, as it is used during resampling, which is currently a part of encode(), so there's no need to keep slices in encode() output.

Tocket: CVS-167405

[Copilot]

Pull Request Overview

This PR removes the slices tensor and its size from the EncodedImage struct, replacing them with a lightweight slices_shape, and updates related functions to pass slices and target_size through as parameters rather than as part of the struct.

• Remove ov::Tensor slices and ImageSize slices_size from EncodedImage
• Add ov::Shape slices_shape to EncodedImage
• Update llava_image_embed_make_with_bytes_slice, encode, and resample_encoded_image to forward slices and target_size, and populate slices_shape

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
src/cpp/src/visual_language/vision_encoder.hpp	Removed slices/slices_size, added slices_shape field
src/cpp/src/visual_language/minicpm/classes.hpp	Updated resample_encoded_image signature to accept slices and target_sizes
src/cpp/src/visual_language/minicpm/classes.cpp	Extended llava_image_embed_make_with_bytes_slice signature and body; updated encode and resample_encoded_image to use new parameters and set slices_shape; updated downstream slice-logic to use slices_shape

Comments suppressed due to low confidence (1)

src/cpp/src/visual_language/minicpm/classes.hpp:41

• [nitpick] Rename parameter target_sizes to target_size to match its singular usage and improve readability.

ResampledImage resample_encoded_image(const EncodedImage& image, const ov::Tensor slices, const ImageSize& target_sizes);

[Copilot]

Pull Request Overview

This PR removes the actual slice tensors from EncodedImage and instead carries only their shape metadata, refactoring downstream resampling to receive slice data via a new helper struct.

• Deleted ov::Tensor slices and ImageSize slices_size, added ov::Shape slices_shape in EncodedImage.
• Introduced ImageSliceResult to bundle slice tensor and target size, and updated llava_image_embed_make_with_bytes_slice to return it.
• Changed resample_encoded_image signature and all call sites to accept explicit slice tensor and target size.

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
src/cpp/src/visual_language/vision_encoder.hpp	Removed slices/slices_size, added slices_shape doc.
src/cpp/src/visual_language/minicpm/classes.hpp	Updated resample_encoded_image declaration to take slices + size.
src/cpp/src/visual_language/minicpm/classes.cpp	Added ImageSliceResult, refactored slicing and resample flows.

Comments suppressed due to low confidence (2)

src/cpp/src/visual_language/minicpm/classes.hpp:29

• [nitpick] The parameter name target_sizes is plural but its type is a single ImageSize; consider renaming it to target_size for clarity.

ResampledImage resample_encoded_image(const EncodedImage& image, const ov::Tensor& slices, const ImageSize& target_sizes);

src/cpp/src/visual_language/minicpm/classes.cpp:288

• No unit tests cover the new multi-slice code path or verify that slices_shape is populated and used correctly. Consider adding tests for both single- and multi-slice scenarios.

std::pair<EncodedImage, ImageSliceResult> llava_image_embed_make_with_bytes_slice(clip_ctx& ctx_clip, const ov::Tensor& img, ov::InferRequest& encoder, int max_slice_nums, int scale_resolution, size_t patch_size, bool never_split) {