Dmitry Kurtaev

Results 21 issues of Dmitry Kurtaev

# What does this PR do? This PR introduces Intel OpenVINO backend for inferencing. Port of https://github.com/huggingface/transformers/pull/14203 ## Before submitting - [ ] This PR fixes a typo or improves...

Use standalone OpenVINO package alternatively to https://github.com/huggingface/optimum/pull/45

demo: https://github.com/openvinotoolkit/open_model_zoo/tree/master/demos/mri_reconstruction_demo example: https://github.com/openvinotoolkit/open_model_zoo/pull/2178 - [ ] AsyncPipeline - [x] Add description for inputs and outputs in the README https://github.com/openvinotoolkit/open_model_zoo/pull/2770

good first issue
demo

At this moment, SparseConv is not implemented in OpenVINO but I've published an extension which provides reference implementation of the following layers: * SparseConv * SparseContTranspose * `calculate_grid`. The method...

Hi! Due Scannet requires an agreement for research usage, please add a simple tests which demonstrates a basic launch of `SparseConvUnet`. Need at least inputs shapes filled by random data.

This PR has changes which let to export trained model to ONNX format compatible with OpenCV (PR https://github.com/opencv/opencv/pull/16925 is required) ### Convert to ONNX ``` python3 export_to_onnx.py --model ssd300_mAP_77.43_v2.pth ```...

We are not taking pull requests at this time.

Optimized version gives me about 1 minute less processing time for StarDist on OS-3.ndpi Proposed to use histogram calculation to compute percentiles. Valid only for RGB images (`[0-256)` pixels range).

enhancement

There is an error with https://github.com/huggingface/transformers/blob/master/examples/pytorch/token-classification/run_ner.py example on `transformers==4.16.1`: ``` [WARNING|trainer_callback.py:304] 2022-01-31 06:54:46,782 >> The Trainer will not work properly if you don't have a `DefaultFlowCallback` in its callbacks. You...

Working on https://github.com/dkurt/optimum-openvino/ found a moment that NNCF provides an option to restore training from compression state: https://github.com/openvinotoolkit/nncf/blob/ecc97dd130d82602e8a4af221b15c1b9d7857dd6/third_party_integration/huggingface_transformers/0001-Modifications-for-NNCF-usage.patch#L1157-L1172 but there is no call of `save_pretrained` with `nncf_compression_state` option

bug