upload videomae-large-camera-motion

[VideoMAE model](https://huggingface.co/docs/transformers/model_doc/videomae)(`large`) variant that has been finetuned for multi-label video classification (a video can belong to multiple classes simultaneously) for camera motion classification on internal dataset.

The model predicts `18` different camera motion `'arc_left', 'arc_right', 'dolly_in', 'dolly_out', 'pan_left', 'pan_right', 'pedestal_down', 'pedestal_up', 'roll_left', 'roll_right', 'static', 'tilt_down', 'tilt_up', 'truck_left', 'truck_right', 'undefined', 'zoom_in', 'zoom_out'` and and 3 shot type classes: `'pov', 'shake', 'track' ` .

Model was trained to associate **entire video** with camera labels, not frame-level motions(!): [input video] -> label/labels (because multilabel) for all video. So, if this camera motion exists during all video frames model should predict this motion, otherwise it should predict `undefined`.

The model is configured to process `config.num_frames=16` frames per input clip. These frames are extracted uniformly from the input video, regardless of its original duration. For videos longer than **2 seconds**, processing the entire video as a single clip may miss temporal nuances (e.g., varying camera motions). So, recommended workflow for such videos will be follows:

(a) split the video into non-overlapping 2-second segments (or sliding windows with optional overlap).

(b) run inference independently on each segment.

(c) post-process results.

config.json

@@ -0,0 +1,77 @@
+{
+  "_name_or_path": "MCG-NJU/videomae-large",
+  "architectures": [
+    "VideoMAEForVideoClassification"
+  ],
+  "attention_probs_dropout_prob": 0.0,
+  "decoder_hidden_size": 512,
+  "decoder_intermediate_size": 2048,
+  "decoder_num_attention_heads": 8,
+  "decoder_num_hidden_layers": 12,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.0,
+  "hidden_size": 1024,
+  "id2label": {
+    "0": "arc_left",
+    "1": "arc_right",
+    "2": "dolly_in",
+    "3": "dolly_out",
+    "4": "pan_left",
+    "5": "pan_right",
+    "6": "pedestal_down",
+    "7": "pedestal_up",
+    "8": "pov",
+    "9": "roll_left",
+    "10": "roll_right",
+    "11": "shake",
+    "12": "static",
+    "13": "tilt_down",
+    "14": "tilt_up",
+    "15": "track",
+    "16": "truck_left",
+    "17": "truck_right",
+    "18": "undefined",
+    "19": "zoom_in",
+    "20": "zoom_out"
+  },
+  "image_size": 224,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "label2id": {
+    "arc_left": 0,
+    "arc_right": 1,
+    "dolly_in": 2,
+    "dolly_out": 3,
+    "pan_left": 4,
+    "pan_right": 5,
+    "pedestal_down": 6,
+    "pedestal_up": 7,
+    "pov": 8,
+    "roll_left": 9,
+    "roll_right": 10,
+    "shake": 11,
+    "static": 12,
+    "tilt_down": 13,
+    "tilt_up": 14,
+    "track": 15,
+    "truck_left": 16,
+    "truck_right": 17,
+    "undefined": 18,
+    "zoom_in": 19,
+    "zoom_out": 20
+  },
+  "layer_norm_eps": 1e-12,
+  "model_type": "videomae",
+  "norm_pix_loss": true,
+  "num_attention_heads": 16,
+  "num_channels": 3,
+  "num_frames": 16,
+  "num_hidden_layers": 24,
+  "patch_size": 16,
+  "problem_type": "multi_label_classification",
+  "qkv_bias": true,
+  "torch_dtype": "float32",
+  "transformers_version": "4.48.3",
+  "tubelet_size": 2,
+  "use_mean_pooling": false
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f76ada1a8fa74170de66e8fc797aca2f448d0ed44ebd8cd9b8106a25803d6ae
+size 1215574156

preprocessor_config.json

@@ -0,0 +1,26 @@
+{
+  "crop_size": {
+    "height": 224,
+    "width": 224
+  },
+  "do_center_crop": true,
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "image_mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "image_processor_type": "VideoMAEImageProcessor",
+  "image_std": [
+    0.229,
+    0.224,
+    0.225
+  ],
+  "resample": 2,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "shortest_edge": 224
+  }
+}