update model and dataset labels

backend/app.py

@@ -1,3 +1,9 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
 import json
 import logging
 import mimetypes

backend/config.py

@@ -1,3 +1,9 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
 # Change these values to match your dataset structure if loading locally or from a different source.
 # IMPORTANT: When running from docker more setup is required (e.g. on Huggingface)
 import os
@@ -130,11 +136,11 @@ MODALITY_CONFIG_CONSTANTS = {
 }
 
 DATASET_CONFIGS = {
-    "voxpopuli_1k/audio": {"type": "audio", "path": ABS_DATASET_PATH},
-    "ravdess_1k/audio": {"type": "audio", "path": ABS_DATASET_PATH},
-    "val2014_1k_v3/image": {"type": "image", "path": ABS_DATASET_PATH},
-    "sa_1b_val_1k/image": {"type": "image", "path": ABS_DATASET_PATH},
-    "sav_val_full_v2/video": {"type": "video", "path": ABS_DATASET_PATH},
+    "Voxpopuli": {"type": "audio", "path": ABS_DATASET_PATH},
+    "Ravdess": {"type": "audio", "path": ABS_DATASET_PATH},
+    "MS-COCO": {"type": "image", "path": ABS_DATASET_PATH},
+    "SA-1B": {"type": "image", "path": ABS_DATASET_PATH},
+    "SA-V": {"type": "video", "path": ABS_DATASET_PATH},
 }
 
 
@@ -213,11 +219,11 @@ def _get_db_key_for_dataset(dataset_base_name):
     """Helper function to determine the database key for a dataset"""
     # Map of dataset names to their db keys
     db_key_mapping = {
-        "voxpopuli_1k": "voxpopuli",
-        "val2014_1k_v3": "local_val2014",
-        "sa_1b_val_1k": "local_valid",
-        "sav_val_full_v2": "sa-v_sav_val_videos",
-        "ravdess_1k": "ravdess",  # Add mapping for ravdess dataset
+        "Voxpopuli": "voxpopuli",
+        "MS-COCO": "local_val2014",
+        "SA-1B": "local_valid",
+        "SA-V": "sa-v_sav_val_videos",
+        "Ravdess": "ravdess",  # Add mapping for ravdess dataset
     }
 
     return db_key_mapping.get(dataset_base_name, dataset_base_name)

backend/descriptions.py

@@ -1,3 +1,9 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
 DESCRIPTIONS = {
     "leaderboard_header": {
         "description": "A leaderboard for evaluating the performance of watermarking methods using the OmniSealBench framework.",
@@ -186,24 +192,24 @@ MODEL_DESCRIPTIONS = {
     "alias": "MBRS"
   },
   "videoseal_0.0": {
-    "full_name": "VideoSeal",
+    "full_name": "VideoSeal (Legacy)",
     "description": "A neural video watermarking system designed to embed imperceptible watermarks that are robust against common video manipulations and processing operations. Legacy model with more robust but visible watermarks.",
     "paper_link": "https://arxiv.org/abs/2412.09492",
     "github_link": "https://github.com/facebookresearch/videoseal",
     "message_size": "96 bits",
-    "alias": "VideoSeal v1.0"
+    "alias": "VideoSeal v0.0"
   },
   "videoseal_1.0": {
-    "full_name": "VideoSeal",
+    "full_name": "VideoSeal v1.0",
     "description": "A neural video watermarking system designed to embed imperceptible watermarks that are robust against common video manipulations and processing operations. Updated model with best balance of efficiency and robustness.",
     "paper_link": "https://arxiv.org/abs/2412.09492",
     "github_link": "https://github.com/facebookresearch/videoseal",
     "message_size": "256 bits",
-    "alias": "VideoSeal v2.0"
+    "alias": "VideoSeal v1.0"
   },
-"pixelseal": {
+  "pixelseal": {
     "full_name": "PixelSeal",
-    "description": "Pixel Seal: Adversarial-only training for invisible image and video watermarking, with improved robustness-imperceptibility tradeoff.",
+    "description": "Adversarial-only training for invisible image and video watermarking.",
     "paper_link": "https://arxiv.org/abs/2412.09492",
     "github_link": "https://github.com/facebookresearch/videoseal",
     "message_size": "256 bits",
@@ -220,35 +226,35 @@ MODEL_DESCRIPTIONS = {
 }
 
 DATASET_DESCRIPTIONS = {
-    "ravdess_1k/audio": {
+    "Ravdess": {
         "full_name": "RAVDESS Emotional speech audio",
         "description": "The RAVDESS dataset contains emotional speech and song recordings, which can be used for audio watermarking tasks. The 1K version includes 1000 samples.",
         "paper_link": "https://doi.org/10.5281/zenodo.1188976",
         "github_link": "",
     },
 
-    "voxpopuli_1k/audio": {
+    "Voxpopuli": {
         "full_name": "VoxPopuli 1K Audio",
         "description": "The VoxPopuli dataset is a large collection of audio recordings from various speakers, suitable for audio watermarking tasks. The 1K version includes 1000 samples.",
         "paper_link": "https://arxiv.org/abs/2101.00390",
         "github_link": "",
     },
 
-    "val2014_1k_v3/image": {
+    "MS-COCO": {
         "full_name": "COCO 2014 Validation Set",
         "description": "The COCO 2014 validation set is a widely used dataset for image watermarking tasks. It contains a diverse set of images with various objects and scenes.",
         "paper_link": "https://arxiv.org/abs/1405.0312",
         "github_link": "",
     },
 
-    "sa_1b_val_1k/image": {
+    "SA-1B": {
         "full_name": "Segment Anything 1 Billion",
         "description": "Segment Anything 1 Billion (SA-1B) is a dataset designed for training general-purpose object segmentation models from open world images.",
         "paper_link": "https://arxiv.org/abs/2304.02643",
         "github_link": "",
     },
 
-    "sav_val_full_v2/video": {
+    "SA-V": {
         "full_name": "SA-Video Dataset",
         "description": "The SA-Video dataset is a collection of videos designed for video watermarking tasks. It includes a variety of video content suitable for testing watermarking techniques.",
         "paper_link": "https://arxiv.org/abs/2401.17264",

backend/examples.py

@@ -1,3 +1,9 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
 import ast
 import json
 import re
@@ -249,6 +255,6 @@ def get_examples_tab(datatype: str, dataset_name: str):
         config["path"],
         datatype=datatype,
         dataset_name=config["dataset_name"],
-        db_key=config["db_key"],
+        db_key=config.get("db_key", ""),
     )
     return infos

backend/tools.py

@@ -1,3 +1,9 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
 import collections
 
 import pandas as pd