Add id2 label file (#4025)

* Add kinetics400-id2label.json

Add kinetics 400 id2 label file

* Update kinetics data specification link

Update to working link

Author: Aseiel

models/intel/action-recognition-0001/README.md

@@ -3,7 +3,7 @@
 ## Use Case and High-Level Description
 
 This is a general-purpose action recognition composite model, consisting of encoder and decoder parts, trained on Kinetics-400 dataset. The encoder model uses Video Transformer approach with ResNet34 encoder.
-Please refer to the [kinetics](https://deepmind.com/research/open-source/open-source-datasets/kinetics/) dataset specification to see list of action that are recognised by this composite model.
+Please refer to the [kinetics](https://research.google/pubs/the-kinetics-human-action-video-dataset/) dataset specification to see list of action that are recognised by this composite model.
 
 ## Example