Update GRDR release checkpoints

Promote latest GRDR checkpoints to canonical best_model paths for msrvtt, actnet, didemo, and panda. X-Pool files are unchanged.

.gitattributes

@@ -72,3 +72,7 @@ GRDR/msrvtt/best_model/best_model.pt.centroids filter=lfs diff=lfs merge=lfs -te
 GRDR/msrvtt/best_model/best_model.pt.code filter=lfs diff=lfs merge=lfs -text
 GRDR/msrvtt/best_model/best_model.pt.embedding filter=lfs diff=lfs merge=lfs -text
 GRDR/msrvtt/best_model/best_model.pt.videorqvae filter=lfs diff=lfs merge=lfs -text
+GRDR/panda/best_model/best_model.pt.centroids filter=lfs diff=lfs merge=lfs -text
+GRDR/panda/best_model/best_model.pt.code filter=lfs diff=lfs merge=lfs -text
+GRDR/panda/best_model/best_model.pt.embedding filter=lfs diff=lfs merge=lfs -text
+GRDR/panda/best_model/best_model.pt.videorqvae filter=lfs diff=lfs merge=lfs -text

GRDR/actnet/best_model/best_model.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7815a29ff7b3df0d7870a8feb122cfd85048b7c29ac8869546ad0ba73ad8fee7
+oid sha256:8064211d28b19185e8d450d9be96a6481b12b63fda2f0e9e77488bd364e55ee1
 size 251294550

GRDR/actnet/best_model/best_model.pt.centroids

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:78b758b80f6a30ff7617b979ed0f9c92e76d3f12d7a63803ed54d6e2eea6adae
+oid sha256:8101db493591eedaf055d5de12c1ab83cdc02182947a8f13ba64e341d0bafae9
 size 788414

GRDR/actnet/best_model/best_model.pt.code

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ea3538673735ca5a19e1c129870422bfa1d723eaeb800af9509855f8b684a85
-size 17176455
+oid sha256:3914bb3978abb2d3f976525e0385258413f38d94eac832979453b89b38aecefc
+size 1246276

GRDR/actnet/best_model/best_model.pt.embedding

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a60c6cff9f8fbc80b7e95b3ca67a7e3da31ca6c33e73d18079c51d3f03e5a027
+oid sha256:f99ddd8f6aef3a0f811e1ffb3926434e2b1d7820e52ffe65495fac1fd6d31053
 size 788350

GRDR/actnet/best_model/best_model.pt.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6593aa8698e0db893674de1dbde2fd1bfeaaf5443ef2f411bc61ef3e06c35b99
+oid sha256:d6adf736fc854b242ec06f880f12ca5b8c04b352cbf9ff1241b56f663e0b050a
 size 242088214

GRDR/actnet/best_model/best_model.pt.videorqvae

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8f97ff998e37ca6ff2da614d55ab53abf8f10308c54119a22312c915c5e0a0a2
+oid sha256:2d9f9c942125c311221bf38913dc5a00772030b694caa035f415ad1c69c219e5
 size 9199783

GRDR/didemo/best_model/best_model.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:100632f58f2c1acead2e3a7119331182455a0f8772754dba6d3a76221c4a4021
-size 251097942
+oid sha256:42d86895b39df148712a444d8e2a2708dc6868e2b7d90f6a10afd076b042da84
+size 251294550

GRDR/didemo/best_model/best_model.pt.centroids

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:125a4f8d55509ef1b791baca9ad1d1e620ceb3150d3c307ae3eab7056770aa29
-size 591806
+oid sha256:642ff4b200e2121dc9a01f7a5db487ecbff3952f7cf38ea7e1c49cf5626e4e81
+size 788414

GRDR/didemo/best_model/best_model.pt.code

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a6c6635286836b92c5f53953aab55edd0021137b1d7a7c802f528cf29ef5931f
-size 24152749
+oid sha256:1b39c57e869570ca2326c004e6b42ab3618162a327851aebfac71c71611e7742
+size 1796059

GRDR/didemo/best_model/best_model.pt.embedding

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4024ac8dcfcdf07c396436b4df958a41f4a511617b26fabf99c14a36e0d03045
-size 591742
+oid sha256:3b53f6ce630a1c459a27cbf5d991ead937613215e9c6d24d790477944cd995c3
+size 788350

GRDR/didemo/best_model/best_model.pt.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:90cccddc1d2f9c00811aa34065dec01697ba472899c36a70d3c8876a1adf0507
+oid sha256:6007cf448db2cf38e83a9d1d3eaa73e75df8986f2a005fa564116af0019b0ee4
 size 242088214

GRDR/didemo/best_model/best_model.pt.videorqvae

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e964bda1571410d2507d8a6ed2f5d6a2497f65bc3265b7024d2e7c928c95d6ea
-size 9003175
+oid sha256:9f351f8f446a407af2304c8b08867154bc3767cf325b2eeb7e6a7b63b27717e6
+size 9199783

GRDR/msrvtt/best_model/best_model.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:08fbb9ba65bc656a29d62b675d775ca8e605a27831fc7ebda01036afc1f34d6c
+oid sha256:be42f8c65dc3e868a03a22b995b872ec613ab0865f4ae1baaf3374f6ec1da8bf
 size 251294550

GRDR/msrvtt/best_model/best_model.pt.centroids

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9fc1275c2ba171ce1de60358d69599d3a18a499a0a0f941531da2cffc624c8c0
+oid sha256:d915f1fc308fc024fa6983e49e8a8e2d0258ef0b3a91aa9fd82187b64590b369
 size 788414

GRDR/msrvtt/best_model/best_model.pt.code

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b60a20d59b876bd7cc154de55df95921982e8700c35550e9213a6cd3db375823
-size 13502130
+oid sha256:4f11eec3011bfecd740e05359fc7bee6b3e50ced345ea9b48b802e52568cdd99
+size 5368953

GRDR/msrvtt/best_model/best_model.pt.embedding

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9efbfdcafb4e9594022a957a38c12c32e25384f110c254eada0424b11b3a29e0
+oid sha256:e7cfdfa6029e6e3eabd38c5f1ff807f1cbcca0560cc0206bc2281971faae86b6
 size 788350

GRDR/msrvtt/best_model/best_model.pt.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eff11e9246cc33dbcc14627c2ade633ffce88b922938951b83425228bd2f39ac
+oid sha256:46cf8b702ebf808249a2f9ec6b40e6aa684e8d622f394e64a7faeaa8fa4e8a51
 size 242088214

GRDR/msrvtt/best_model/best_model.pt.videorqvae

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d95d54b2ae5789cb715c787a9548f4f6d9492065f64d7eb5c464c97ed3a99f15
+oid sha256:95eb0ad22e74ef00e21e23c1734f4e11ea66015a7f911539544bccbab88f3311
 size 9199783

GRDR/panda/best_model/best_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60312639d96c325bf9b01d03bd2db3a0767d05bcd239b291db25849ceaf6f828
+size 275674345

GRDR/panda/best_model/best_model.pt.centroids

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c3aba66ca0bc253132083caff489b1e106e8a1ac37c430c197e712cd30242f3
+size 25168279

GRDR/panda/best_model/best_model.pt.code

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3cac0c18d7703512e42afd6214ca8698f6c6f34f2f9a7853e288bae94d4cda6
+size 561266683

GRDR/panda/best_model/best_model.pt.embedding

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8c5d535f5ad4889668cdc834d51be15d16488eae209e325c34d0d91e9918a45
+size 25168151

GRDR/panda/best_model/best_model.pt.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22578e363d467f5be3b001beda172319d7cf99d16cf6b9dc64a8a1808d6ad072
+size 242088623

GRDR/panda/best_model/best_model.pt.videorqvae

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8622641b64db5ed8abfa21385bd86d7f8e96469479617bf6a878d5641316d69c
+size 33579584

README.md

@@ -14,118 +14,78 @@ size_categories:
 - 10K<n<100K
 ---
 
-# GRDR-TVR: Generative Recall, Dense Reranking for Text-to-Video Retrieval
+# GRDR-TVR release assets
 
-This dataset contains the pre-extracted video features and trained model checkpoints for the GRDR (Generative Recall, Dense Reranking) framework for efficient Text-to-Video Retrieval (TVR).
+This dataset repository hosts release assets for GRDR: a learned semantic-key index for large-scale text-to-video retrieval. The main code repository is [JasonCoderMaker/GRDR](https://github.com/JasonCoderMaker/GRDR).
 
-## 📄 Paper
+## Hosted files
 
-**Generative Recall, Dense Reranking: Learning Multi-View Semantic IDs for Efficient Text-to-Video Retrieval**
+### InternVideo2 features
 
+Pre-extracted InternVideo2 features are available for:
 
-[Paper PDF](https://arxiv.org/abs/XXXX.XXXXX) | [Code Repository](https://github.com/JasonCoderMaker/GRDR)
+- `msrvtt`
+- `actnet`
+- `didemo`
+- `lsmdc` legacy assets
 
-## 📊 Dataset Overview
+Panda GRDR checkpoints are hosted here, but Panda feature and annotation preparation follows the code repository instructions.
 
-This dataset includes three main components:
+### GRDR checkpoints
 
-### 1. InternVideo2 Features (~3.4GB)
-Pre-extracted video features using InternVideo2 encoder for four benchmark datasets:
-- **MSR-VTT**: 10,000 videos (932MB)
-- **ActivityNet**: 20,000 videos (1.1GB)
-- **DiDeMo**: 10,464 videos (916MB)
-- **LSMDC**: 1,000 movies, 118,081 clips (424MB)
+The release-facing checkpoint path is canonical for each dataset:
 
-**Feature Details:**
-- Dimension: 512-d embeddings
-- Format: Pickle files (`.pkl`) with `{video_id: embedding}` mappings
-- Extraction: InternVideo2 (InternVL-2B) with temporal pooling
+```text
+GRDR/<dataset>/best_model/best_model.pt
+GRDR/<dataset>/best_model/best_model.pt.code
+GRDR/<dataset>/best_model/best_model.pt.centroids
+GRDR/<dataset>/best_model/best_model.pt.embedding
+GRDR/<dataset>/best_model/best_model.pt.model
+GRDR/<dataset>/best_model/best_model.pt.start_token
+GRDR/<dataset>/best_model/best_model.pt.videorqvae
+```
 
-### 2. GRDR Model Checkpoints (~2GB)
-Trained GRDR models (T5-small based) for all four datasets:
-- **MSR-VTT**: 494MB
-- **ActivityNet**: 498MB  
-- **DiDeMo**: 504MB
-- **LSMDC**: 478MB
+Current release datasets:
 
-**Checkpoint Components:**
-- `best_model.pt` - Complete model checkpoint
-- `best_model.pt.model` - T5 encoder-decoder weights
-- `best_model.pt.videorqvae` - Video RQ-VAE quantizer
-- `best_model.pt.code` - Pre-computed semantic IDs
-- `best_model.pt.centroids` - Codebook centroids
-- `best_model.pt.embedding` - Learned embeddings
-- `best_model.pt.start_token` - Start token embeddings
+- `msrvtt`
+- `actnet`
+- `didemo`
+- `panda`
 
-**Model Architecture:**
-- Base: T5-small (60M parameters)
-- Codebook size: 128/96/200 (dataset-dependent)
-- Max code length: 3
-- Training: 3-phase progressive training
+Legacy `lsmdc` checkpoint files remain available under `GRDR/lsmdc/best_model/`, but they are not part of the current GRDR release manifest.
 
-### 3. Xpool Reranker Checkpoints (~7.2GB)
-Pre-trained reranker models for dense reranking stage:
-- **MSR-VTT**: msrvtt9k_model_best.pth (1.8GB)
-- **ActivityNet**: actnet_model_best.pth (1.8GB)
-- **DiDeMo**: didemo_model_best.pth (1.8GB)
-- **LSMDC**: lsmdc_model_best.pth (1.8GB)
+### X-Pool checkpoints
 
-### 4. Xpool Video Features (~3.2GB)
-Pre-extracted CLIP video features for Xpool reranker:
-- **MSR-VTT**: 235MB
-- **ActivityNet**: 351MB
-- **DiDeMo**: 221MB
-- **LSMDC**: 2.4GB
+X-Pool checkpoint filenames are kept in their original flat layout:
 
-**Reranker Details:**
-- Architecture: CLIP-based (ViT-B/32)
-- Purpose: Fine-grained reranking of recalled candidates
-- Format: PyTorch checkpoint files (`.pth`)
+```text
+Xpool/msrvtt9k_model_best.pth
+Xpool/actnet_model_best.pth
+Xpool/didemo_model_best.pth
+Xpool/lsmdc_model_best.pth
+```
 
-## 📂 Repository Structure
+No X-Pool checkpoint file was renamed for this release update.
 
-```
-GRDR-TVR/
-├── README.md                          # This file
-├── download_features.py               # Python download utility
-├── download_checkpoints.sh            # Bash download script
-│
-├── InternVideo2/                      # Video Features (3.4GB)
-│   ├── actnet/
-│   │   └── actnet_internvideo2.pkl
-│   ├── didemo/
-│   │   └── didemo_internvideo2.pkl
-│   ├── lsmdc/
-│   │   └── lsmdc_internvideo2.pkl
-│   └── msrvtt/
-│       └── msrvtt_internvideo2.pkl
-│
-├── GRDR/                              # GRDR Checkpoints (2GB)
-│   ├── actnet/best_model/
-│   ├── didemo/best_model/
-│   ├── lsmdc/best_model/
-│   └── msrvtt/best_model/
-│
-└── Xpool/                             # Reranker Checkpoints (7.2GB)
-    ├── actnet_model_best.pth
-    ├── didemo_model_best.pth
-    ├── lsmdc_model_best.pth
-    └── msrvtt9k_model_best.pth
-```
+## Download examples
 
-## 📝 License
+Use the `download_features.py` script from the GRDR code repository for the current release workflow:
 
-This dataset is released under the MIT License. See [LICENSE](LICENSE) for details.
+```bash
+python download_features.py --grdr --datasets msrvtt panda
+python download_features.py --features --datasets msrvtt actnet didemo
+python download_features.py --xpool --datasets msrvtt actnet didemo
+```
 
-The video datasets (MSR-VTT, ActivityNet, DiDeMo, LSMDC) are subject to their original licenses. This repository only provides pre-extracted features, not the original videos.
+You can also download directly from the Hub:
 
-## 🙏 Acknowledgments
+```bash
+hf download JasonCoderMaker/GRDR-TVR --type dataset --include 'GRDR/msrvtt/best_model/*' --local-dir ./output/checkpoints
+hf download JasonCoderMaker/GRDR-TVR --type dataset --include 'Xpool/msrvtt9k_model_best.pth' --local-dir ./reranker/xpool/ckpt
+```
 
-- **InternVideo2**: We thank the authors of InternVideo2 for their excellent video encoder
-- **Xpool**: The reranker architecture is based on X-POOL
-- **Datasets**: MSR-VTT, ActivityNet Captions, DiDeMo, and LSMDC benchmark creators
+The GRDR repository expects checkpoints under `output/checkpoints/GRDR/<dataset>/best_model/`.
 
+## License
 
-**Dataset Version**: 1.0  
-**Last Updated**: January 2026  
-**Maintained by**: [@JasonCoderMaker](https://huggingface.co/JasonCoderMaker)
+The release assets are distributed under the MIT License where applicable. Source video datasets keep their original licenses. This repository provides features and model checkpoints, not the original videos.

download_checkpoints.sh

@@ -1,284 +1,70 @@
-#!/bin/bash
-###############################################################################
-# Download GRDR-TVR Dataset from Hugging Face Hub
-#
-# This script provides a simple way to download the GRDR-TVR dataset
-# components using the Hugging Face CLI.
-#
-# Usage:
-#   ./download_checkpoints.sh [OPTIONS]
-#
-# Examples:
-#   # Download everything
-#   ./download_checkpoints.sh --all
-#
-#   # Download only GRDR checkpoints
-#   ./download_checkpoints.sh --grdr
-#
-#   # Download features for specific dataset
-#   ./download_checkpoints.sh --features msrvtt
-###############################################################################
+#!/usr/bin/env bash
+set -euo pipefail
 
 REPO_ID="JasonCoderMaker/GRDR-TVR"
-OUTPUT_DIR="./GRDR-TVR"
-
-# Colors for output
-RED='\033[0;31m'
-GREEN='\033[0;32m'
-YELLOW='\033[1;33m'
-BLUE='\033[0;34m'
-NC='\033[0m' # No Color
-
-# Print colored message
-print_msg() {
-    local color=$1
-    shift
-    echo -e "${color}$@${NC}"
-}
-
-# Check if huggingface-cli is installed
-check_hf_cli() {
-    if ! command -v huggingface-cli &> /dev/null; then
-        print_msg $RED "Error: huggingface-cli not found!"
-        print_msg $YELLOW "Please install: pip install huggingface_hub"
-        exit 1
-    fi
-}
-
-# Download all components
-download_all() {
-    print_msg $BLUE "=================================================="
-    print_msg $BLUE "Downloading complete GRDR-TVR dataset..."
-    print_msg $BLUE "=================================================="
-    
-    huggingface-cli download $REPO_ID \
-        --repo-type dataset \
-        --local-dir $OUTPUT_DIR \
-        --local-dir-use-symlinks False
-    
-    print_msg $GREEN "✓ Complete dataset downloaded to $OUTPUT_DIR"
-}
-
-# Download InternVideo2 features
-download_features() {
-    local dataset=$1
-    print_msg $BLUE "=================================================="
-    print_msg $BLUE "Downloading InternVideo2 features..."
-    print_msg $BLUE "=================================================="
-    
-    if [ -z "$dataset" ]; then
-        # Download all features
-        huggingface-cli download $REPO_ID \
-            --repo-type dataset \
-            --include "InternVideo2/*" \
-            --local-dir $OUTPUT_DIR \
-            --local-dir-use-symlinks False
-    else
-        # Download specific dataset features
-        huggingface-cli download $REPO_ID \
-            --repo-type dataset \
-            --include "InternVideo2/${dataset}/*" \
-            --local-dir $OUTPUT_DIR \
-            --local-dir-use-symlinks False
-    fi
-    
-    print_msg $GREEN "✓ Features downloaded to $OUTPUT_DIR/InternVideo2"
-}
-
-# Download GRDR checkpoints
-download_grdr() {
-    local dataset=$1
-    print_msg $BLUE "=================================================="
-    print_msg $BLUE "Downloading GRDR checkpoints..."
-    print_msg $BLUE "=================================================="
-    
-    if [ -z "$dataset" ]; then
-        # Download all GRDR checkpoints
-        huggingface-cli download $REPO_ID \
-            --repo-type dataset \
-            --include "GRDR/*" \
-            --local-dir $OUTPUT_DIR \
-            --local-dir-use-symlinks False
-    else
-        # Download specific dataset checkpoint
-        huggingface-cli download $REPO_ID \
-            --repo-type dataset \
-            --include "GRDR/${dataset}/*" \
-            --local-dir $OUTPUT_DIR \
-            --local-dir-use-symlinks False
-    fi
-    
-    print_msg $GREEN "✓ GRDR checkpoints downloaded to $OUTPUT_DIR/GRDR"
-}
-
-# Download Xpool checkpoints
-download_xpool() {
-    local dataset=$1
-    print_msg $BLUE "=================================================="
-    print_msg $BLUE "Downloading Xpool reranker checkpoints..."
-    print_msg $BLUE "=================================================="
-    
-    if [ -z "$dataset" ]; then
-        # Download all Xpool checkpoints
-        huggingface-cli download $REPO_ID \
-            --repo-type dataset \
-            --include "Xpool/*" \
-            --local-dir $OUTPUT_DIR \
-            --local-dir-use-symlinks False
-    else
-        # Download specific dataset checkpoint
-        local filename=""
-        case $dataset in
-            msrvtt)
-                filename="msrvtt9k_model_best.pth"
-                ;;
-            actnet)
-                filename="actnet_model_best.pth"
-                ;;
-            didemo)
-                filename="didemo_model_best.pth"
-                ;;
-            lsmdc)
-                filename="lsmdc_model_best.pth"
-                ;;
-            *)
-                print_msg $RED "Error: Unknown dataset '$dataset'"
-                exit 1
-                ;;
-        esac
-        
-        huggingface-cli download $REPO_ID \
-            --repo-type dataset \
-            --include "Xpool/${filename}" \
-            --local-dir $OUTPUT_DIR \
-            --local-dir-use-symlinks False
-    fi
-    
-    print_msg $GREEN "✓ Xpool checkpoints downloaded to $OUTPUT_DIR/Xpool"
-}
-
-# Show usage
-show_usage() {
-    cat << EOF
-Usage: $0 [OPTIONS] [DATASET]
-
-Download GRDR-TVR dataset components from Hugging Face Hub.
-
-OPTIONS:
-    --all           Download all components (default if no option specified)
-    --features      Download InternVideo2 features only
-    --grdr          Download GRDR checkpoints only
-    --xpool         Download Xpool reranker checkpoints only
-    -o, --output    Specify output directory (default: ./GRDR-TVR)
-    -h, --help      Show this help message
-
-DATASET (optional):
-    msrvtt          MSR-VTT dataset
-    actnet          ActivityNet dataset
-    didemo          DiDeMo dataset
-    lsmdc           LSMDC dataset
-    
-    If not specified, downloads all datasets.
-
-EXAMPLES:
-    # Download everything
-    $0 --all
-
-    # Download only GRDR checkpoints
-    $0 --grdr
-
-    # Download InternVideo2 features for MSR-VTT
-    $0 --features msrvtt
-
-    # Download Xpool reranker for ActivityNet
-    $0 --xpool actnet
-
-    # Download to custom directory
-    $0 --all -o ./my_data
-
+OUTPUT_DIR="${OUTPUT_DIR:-./GRDR-TVR}"
+DATASET="${2:-}"
+
+usage() {
+    cat <<'EOF'
+Usage: ./download_checkpoints.sh [--all|--features|--grdr|--xpool] [dataset]
+
+Datasets: msrvtt, actnet, didemo, panda, lsmdc
+Examples:
+  ./download_checkpoints.sh --grdr msrvtt
+  ./download_checkpoints.sh --grdr panda
+  ./download_checkpoints.sh --xpool msrvtt
 EOF
 }
 
-# Main script
-main() {
-    check_hf_cli
-    
-    # Parse arguments
-    local mode="all"
-    local dataset=""
-    
-    while [[ $# -gt 0 ]]; do
-        case $1 in
-            --all)
-                mode="all"
-                shift
-                ;;
-            --features)
-                mode="features"
-                shift
-                ;;
-            --grdr)
-                mode="grdr"
-                shift
-                ;;
-            --xpool)
-                mode="xpool"
-                shift
-                ;;
-            -o|--output)
-                OUTPUT_DIR="$2"
-                shift 2
-                ;;
-            -h|--help)
-                show_usage
-                exit 0
-                ;;
-            msrvtt|actnet|didemo|lsmdc)
-                dataset="$1"
-                shift
-                ;;
-            *)
-                print_msg $RED "Error: Unknown option '$1'"
-                show_usage
-                exit 1
-                ;;
-        esac
-    done
-    
-    print_msg $BLUE "=================================================="
-    print_msg $BLUE "GRDR-TVR Dataset Downloader"
-    print_msg $BLUE "=================================================="
-    print_msg $YELLOW "Repository: $REPO_ID"
-    print_msg $YELLOW "Output: $OUTPUT_DIR"
-    if [ -n "$dataset" ]; then
-        print_msg $YELLOW "Dataset: $dataset"
-    fi
-    print_msg $BLUE "==================================================\n"
-    
-    # Execute download based on mode
-    case $mode in
-        all)
-            download_all
-            ;;
-        features)
-            download_features "$dataset"
-            ;;
-        grdr)
-            download_grdr "$dataset"
-            ;;
-        xpool)
-            download_xpool "$dataset"
-            ;;
-    esac
-    
-    print_msg $GREEN "\n=================================================="
-    print_msg $GREEN "✓ Download Complete!"
-    print_msg $GREEN "=================================================="
-    print_msg $YELLOW "\nNext steps:"
-    print_msg $YELLOW "1. Verify downloads in $OUTPUT_DIR"
-    print_msg $YELLOW "2. See README.md for usage instructions"
-    print_msg $YELLOW "3. Visit https://huggingface.co/datasets/$REPO_ID\n"
+hf_download() {
+    hf download "$REPO_ID" --type dataset --include "$1" --local-dir "$OUTPUT_DIR"
 }
 
-# Run main function
-main "$@"
+cmd="${1:---all}"
+case "$cmd" in
+    --all)
+        if [[ -n "$DATASET" ]]; then
+            hf_download "GRDR/${DATASET}/best_model/*"
+            hf_download "InternVideo2/${DATASET}/**" || true
+            case "$DATASET" in
+                msrvtt) hf_download "Xpool/msrvtt9k_model_best.pth" ;;
+                actnet) hf_download "Xpool/actnet_model_best.pth" ;;
+                didemo) hf_download "Xpool/didemo_model_best.pth" ;;
+                lsmdc) hf_download "Xpool/lsmdc_model_best.pth" ;;
+                panda) echo "No Panda X-Pool checkpoint is hosted in this repo" ;;
+            esac
+        else
+            hf_download "GRDR/*/best_model/*"
+            hf_download "InternVideo2/**"
+            hf_download "Xpool/*"
+        fi
+        ;;
+    --features)
+        [[ -n "$DATASET" ]] && hf_download "InternVideo2/${DATASET}/**" || hf_download "InternVideo2/**"
+        ;;
+    --grdr)
+        [[ -n "$DATASET" ]] && hf_download "GRDR/${DATASET}/best_model/*" || hf_download "GRDR/*/best_model/*"
+        ;;
+    --xpool)
+        if [[ -z "$DATASET" ]]; then
+            hf_download "Xpool/*"
+        else
+            case "$DATASET" in
+                msrvtt) hf_download "Xpool/msrvtt9k_model_best.pth" ;;
+                actnet) hf_download "Xpool/actnet_model_best.pth" ;;
+                didemo) hf_download "Xpool/didemo_model_best.pth" ;;
+                lsmdc) hf_download "Xpool/lsmdc_model_best.pth" ;;
+                panda) echo "No Panda X-Pool checkpoint is hosted in this repo" ;;
+                *) echo "Unknown dataset: $DATASET" >&2; exit 1 ;;
+            esac
+        fi
+        ;;
+    -h|--help)
+        usage
+        ;;
+    *)
+        usage >&2
+        exit 1
+        ;;
+esac

download_features.py

@@ -1,258 +1,99 @@
 #!/usr/bin/env python3
-"""
-Download GRDR-TVR dataset components from Hugging Face Hub.
-
-This script provides a convenient way to download specific components
-of the GRDR-TVR dataset including InternVideo2 features, GRDR checkpoints,
-and Xpool reranker models.
-
-Examples:
-    # Download everything
-    python download_features.py --all
-
-    # Download only features for MSR-VTT and ActivityNet
-    python download_features.py --features --datasets msrvtt actnet
-
-    # Download GRDR checkpoints for all datasets
-    python download_features.py --grdr
-
-    # Download Xpool reranker for specific dataset
-    python download_features.py --xpool --datasets msrvtt
-"""
+"""Download GRDR-TVR release assets from Hugging Face Hub."""
 
 import argparse
-import os
+import shutil
+import tempfile
 from pathlib import Path
-from huggingface_hub import snapshot_download, hf_hub_download
-from tqdm import tqdm
-
 
 REPO_ID = "JasonCoderMaker/GRDR-TVR"
-DATASETS = ["msrvtt", "actnet", "didemo", "lsmdc"]
-
-
-def download_internvideo2_features(datasets, output_dir="./dataset/features"):
-    """Download InternVideo2 pre-extracted features."""
-    print(f"\n{'='*70}")
-    print("📥 Downloading InternVideo2 Features")
-    print(f"{'='*70}\n")
-    
-    features_dir = Path(output_dir) / "InternVideo2"
-    features_dir.mkdir(parents=True, exist_ok=True)
-    
-    for dataset in datasets:
-        print(f"\n📦 Downloading {dataset} features...")
-        try:
-            snapshot_download(
-                repo_id=REPO_ID,
-                repo_type="dataset",
-                allow_patterns=f"InternVideo2/{dataset}/*",
-                local_dir=output_dir,
-                local_dir_use_symlinks=False,
-            )
-            print(f"✓ {dataset} features downloaded to {features_dir / dataset}")
-        except Exception as e:
-            print(f"✗ Error downloading {dataset} features: {e}")
-
-
-def download_grdr_checkpoints(datasets, output_dir="./output"):
-    """Download GRDR model checkpoints."""
-    print(f"\n{'='*70}")
-    print("📥 Downloading GRDR Checkpoints")
-    print(f"{'='*70}\n")
-    
-    grdr_dir = Path(output_dir) / "GRDR"
-    grdr_dir.mkdir(parents=True, exist_ok=True)
-    
-    for dataset in datasets:
-        print(f"\n📦 Downloading {dataset} GRDR checkpoint...")
-        try:
-            snapshot_download(
-                repo_id=REPO_ID,
-                repo_type="dataset",
-                allow_patterns=f"GRDR/{dataset}/**",
-                local_dir=output_dir,
-                local_dir_use_symlinks=False,
-            )
-            print(f"✓ {dataset} GRDR checkpoint downloaded to {grdr_dir / dataset}")
-        except Exception as e:
-            print(f"✗ Error downloading {dataset} GRDR checkpoint: {e}")
-
+DATASETS = ["msrvtt", "actnet", "didemo", "panda", "lsmdc"]
+FEATURE_DATASETS = {"msrvtt", "actnet", "didemo", "lsmdc"}
+GRDR_DATASETS = {"msrvtt", "actnet", "didemo", "panda", "lsmdc"}
+XPOOL_FILES = {
+    "actnet": "actnet_model_best.pth",
+    "didemo": "didemo_model_best.pth",
+    "lsmdc": "lsmdc_model_best.pth",
+    "msrvtt": "msrvtt9k_model_best.pth",
+}
+
+
+def download_features(datasets, output_dir="./dataset/features"):
+    from huggingface_hub import snapshot_download
+
+    selected = [d for d in datasets if d in FEATURE_DATASETS]
+    skipped = sorted(set(datasets) - set(selected))
+    for dataset in skipped:
+        print(f"Skipping {dataset} InternVideo2 features: not hosted in this repo")
+    for dataset in selected:
+        snapshot_download(
+            repo_id=REPO_ID,
+            repo_type="dataset",
+            allow_patterns=f"InternVideo2/{dataset}/**",
+            local_dir=output_dir,
+            local_dir_use_symlinks=False,
+        )
+        print(f"Downloaded {dataset} InternVideo2 features")
+
+
+def download_grdr(datasets, output_dir="./output/checkpoints"):
+    from huggingface_hub import snapshot_download
+
+    selected = [d for d in datasets if d in GRDR_DATASETS]
+    for dataset in selected:
+        snapshot_download(
+            repo_id=REPO_ID,
+            repo_type="dataset",
+            allow_patterns=f"GRDR/{dataset}/best_model/**",
+            local_dir=output_dir,
+            local_dir_use_symlinks=False,
+        )
+        print(f"Downloaded {dataset} GRDR checkpoint")
+
+
+def download_xpool(datasets, output_dir="./reranker/xpool/ckpt"):
+    from huggingface_hub import hf_hub_download
 
-def download_xpool_checkpoints(datasets, output_dir="./reranker/xpool/ckpt"):
-    """Download Xpool reranker checkpoints."""
-    print(f"\n{'='*70}")
-    print("📥 Downloading Xpool Reranker Checkpoints")
-    print(f"{'='*70}\n")
-    
     xpool_dir = Path(output_dir)
     xpool_dir.mkdir(parents=True, exist_ok=True)
-    
-    xpool_files = {
-        "actnet": "actnet_model_best.pth",
-        "didemo": "didemo_model_best.pth",
-        "lsmdc": "lsmdc_model_best.pth",
-        "msrvtt": "msrvtt9k_model_best.pth",
-    }
-    
     for dataset in datasets:
-        if dataset not in xpool_files:
-            print(f"⊘ Skipping {dataset} (no Xpool checkpoint)")
+        filename = XPOOL_FILES.get(dataset)
+        if filename is None:
+            print(f"Skipping {dataset} X-Pool checkpoint: not hosted in this repo")
             continue
-            
-        filename = xpool_files[dataset]
-        print(f"\n📦 Downloading {dataset} Xpool checkpoint...")
-        
-        try:
-            file_path = hf_hub_download(
+        with tempfile.TemporaryDirectory() as tmp:
+            src = hf_hub_download(
                 repo_id=REPO_ID,
                 repo_type="dataset",
                 filename=f"Xpool/{filename}",
-                local_dir=xpool_dir.parent.parent,
+                local_dir=tmp,
                 local_dir_use_symlinks=False,
             )
-            print(f"✓ {dataset} Xpool checkpoint downloaded to {xpool_dir / filename}")
-        except Exception as e:
-            print(f"✗ Error downloading {dataset} Xpool checkpoint: {e}")
-
-
-def download_scripts(output_dir="./scripts"):
-    """Download utility scripts."""
-    print(f"\n{'='*70}")
-    print("📥 Downloading Utility Scripts")
-    print(f"{'='*70}\n")
-    
-    scripts_dir = Path(output_dir)
-    scripts_dir.mkdir(parents=True, exist_ok=True)
-    
-    script_files = [
-        "download_features.py",
-        "download_checkpoints.sh",
-    ]
-    
-    for script in script_files:
-        try:
-            file_path = hf_hub_download(
-                repo_id=REPO_ID,
-                repo_type="dataset",
-                filename=script,
-                local_dir=".",
-                local_dir_use_symlinks=False,
-            )
-            print(f"✓ {script} downloaded")
-        except Exception as e:
-            print(f"⊘ {script} not available: {e}")
+            shutil.copy2(src, xpool_dir / filename)
+        print(f"Downloaded {dataset} X-Pool checkpoint")
 
 
 def main():
-    parser = argparse.ArgumentParser(
-        description="Download GRDR-TVR dataset components from Hugging Face Hub",
-        formatter_class=argparse.RawDescriptionHelpFormatter,
-        epilog="""
-Examples:
-  # Download everything for all datasets
-  python download_features.py --all
-
-  # Download only InternVideo2 features for MSR-VTT
-  python download_features.py --features --datasets msrvtt
+    parser = argparse.ArgumentParser(description="Download GRDR-TVR release assets")
+    parser.add_argument("--all", action="store_true", help="Download hosted features, GRDR checkpoints, and X-Pool checkpoints")
+    parser.add_argument("--features", action="store_true", help="Download hosted InternVideo2 features")
+    parser.add_argument("--grdr", action="store_true", help="Download GRDR checkpoints")
+    parser.add_argument("--xpool", action="store_true", help="Download X-Pool checkpoints")
+    parser.add_argument("--datasets", nargs="+", choices=DATASETS, default=["msrvtt", "actnet", "didemo", "panda"], help="Datasets to download")
+    parser.add_argument("--features-dir", default="./dataset/features")
+    parser.add_argument("--grdr-dir", default="./output/checkpoints")
+    parser.add_argument("--xpool-dir", default="./reranker/xpool/ckpt")
+    args = parser.parse_args()
 
-  # Download GRDR checkpoints for MSR-VTT and ActivityNet
-  python download_features.py --grdr --datasets msrvtt actnet
+    if not any([args.all, args.features, args.grdr, args.xpool]):
+        parser.error("Select at least one component: --all, --features, --grdr, or --xpool")
 
-  # Download all components for DiDeMo
-  python download_features.py --all --datasets didemo
-        """
-    )
-    
-    # Component selection
-    parser.add_argument(
-        "--all",
-        action="store_true",
-        help="Download all components (features, GRDR, Xpool)",
-    )
-    parser.add_argument(
-        "--features",
-        action="store_true",
-        help="Download InternVideo2 features",
-    )
-    parser.add_argument(
-        "--grdr",
-        action="store_true",
-        help="Download GRDR model checkpoints",
-    )
-    parser.add_argument(
-        "--xpool",
-        action="store_true",
-        help="Download Xpool reranker checkpoints",
-    )
-    parser.add_argument(
-        "--scripts",
-        action="store_true",
-        help="Download utility scripts",
-    )
-    
-    # Dataset selection
-    parser.add_argument(
-        "--datasets",
-        nargs="+",
-        choices=DATASETS,
-        default=DATASETS,
-        help="Datasets to download (default: all)",
-    )
-    
-    # Output directories
-    parser.add_argument(
-        "--features-dir",
-        type=str,
-        default="./dataset/features",
-        help="Output directory for features (default: ./dataset/features)",
-    )
-    parser.add_argument(
-        "--grdr-dir",
-        type=str,
-        default="./output",
-        help="Output directory for GRDR checkpoints (default: ./output)",
-    )
-    parser.add_argument(
-        "--xpool-dir",
-        type=str,
-        default="./reranker/xpool/ckpt",
-        help="Output directory for Xpool checkpoints (default: ./reranker/xpool/ckpt)",
-    )
-    
-    args = parser.parse_args()
-    
-    # Validate: at least one component must be selected
-    if not any([args.all, args.features, args.grdr, args.xpool, args.scripts]):
-        parser.error("Please specify at least one component: --all, --features, --grdr, --xpool, or --scripts")
-    
-    print(f"\n{'='*70}")
-    print(f"GRDR-TVR Dataset Downloader")
-    print(f"{'='*70}")
-    print(f"Repository: {REPO_ID}")
-    print(f"Datasets: {', '.join(args.datasets)}")
-    print(f"{'='*70}\n")
-    
-    # Download components
     if args.all or args.features:
-        download_internvideo2_features(args.datasets, args.features_dir)
-    
+        download_features(args.datasets, args.features_dir)
     if args.all or args.grdr:
-        download_grdr_checkpoints(args.datasets, args.grdr_dir)
-    
+        download_grdr(args.datasets, args.grdr_dir)
     if args.all or args.xpool:
-        download_xpool_checkpoints(args.datasets, args.xpool_dir)
-    
-    if args.scripts:
-        download_scripts()
-    
-    print(f"\n{'='*70}")
-    print("✓ Download Complete!")
-    print(f"{'='*70}\n")
-    print("Next steps:")
-    print("1. Verify downloads in the output directories")
-    print("2. See README.md for usage instructions")
-    print(f"3. Visit https://huggingface.co/datasets/{REPO_ID} for more details\n")
+        download_xpool(args.datasets, args.xpool_dir)
 
 
 if __name__ == "__main__":