File size: 2,433 Bytes
b584148 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 | # DROID Preprocessing - Quick Start
## Fast Parallel Processing (Recommended)
### 8 GPUs
```bash
cd /root/workspace/code/wmrl/Dual-Dynamics-Models/DROID-main
bash scripts/run_parallel_8gpu.sh
```
### 16 GPUs
```bash
cd /root/workspace/code/wmrl/Dual-Dynamics-Models/DROID-main
bash scripts/run_parallel_16gpu.sh
```
**How it works:**
- Divides ~70k episodes into equal chunks
- Each GPU processes a continuous range of episodes
- GPU 0: episodes 0-4427
- GPU 1: episodes 4427-8854
- GPU 2: episodes 8854-13281
- etc.
- **Much faster startup** - no dataset iteration overhead
## Two Machines (16 GPUs Total - Optional)
Each machine runs 8 independent GPU processes with no communication.
**Machine 1** (8 GPUs, processes episodes 0,1,2...7,16,17...):
```bash
cd /root/workspace/code/wmrl/Dual-Dynamics-Models/DROID-main
bash scripts/run_multigpu_machine1.sh
```
**Machine 2** (8 GPUs, processes episodes 8,9,10...15,24,25...):
```bash
cd /root/workspace/code/wmrl/Dual-Dynamics-Models/DROID-main
bash scripts/run_multigpu_machine2.sh
```
**How it works:**
- 16 total independent processes (8 per machine)
- No communication between any GPUs or machines
- Episodes distributed via modulo: `episode_idx % 16 == global_gpu_id`
- Both machines write to same shared output directory
## Monitor
```bash
# Watch GPU 0 log
tail -f /mnt/kevin/data/droid_processed_1000pts/log_machine0_gpu0.txt
# Check all processes
ps aux | grep preprocess_droid_multigpu
# GPU utilization
watch -n 1 nvidia-smi
```
## After Completion
```bash
# Count results
ls -1 /mnt/kevin/data/droid_processed_1000pts/data/*.npz | wc -l
# Check preview videos
ls -1 /mnt/kevin/data/droid_processed_1000pts/preview_videos/*.mp4 | wc -l
```
## Stop All
```bash
pkill -f preprocess_droid_rlds_final
```
## Key Info
- **Output**: `/mnt/kevin/data/droid_processed_1000pts/`
- **Tracks per view**: 1000 points (like LIBERO's 1098)
- **Preview videos**: 20 total
- **Processing time**: ~18-24 hours
- **Output size**: ~500GB-1TB
- **Visibility threshold**: 0.7 (in This&That dataloader)
## Requirements
- **CoTracker checkpoint**: Must be at one of these locations:
- `/mnt/kevin/vlm_models/cotracker/scaled_offline.pth` (preferred)
- `/mnt/kevin/vlm_models/hub/checkpoints/scaled_offline.pth`
- `/root/workspace/code/wmrl/Dual-Dynamics-Models/DROID-main/co-tracker/checkpoints/scaled_offline.pth`
See `MULTIGPU_PREPROCESSING_README.md` for full details.
|