============================================================ 🎯 OCULUS EXTENDED DETECTION TRAINER ============================================================ [Loading Model] ✨ Resuming from V2 checkpoint: checkpoints/oculus_detection_v2/final ✓ Loaded projector from checkpoints/oculus_detection_v2/final ✓ Loaded pre-trained detection heads [Oculus] Loading vision encoders... Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.52, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. Loading weights: 0%| | 0/439 [00:00 2048 Progress: 50/100 Mean IoU: 0.1759 Accuracy (IoU>0.5 + correct class): 0.0272 Loaded 0 counting samples 🔢 Counting Benchmark ---------------------------------------- Exact Accuracy: 0.00% Within-1 Accuracy: 0.00% Mean Absolute Error: 0.00 Loaded 0 VQA samples ❓ VQA Benchmark ---------------------------------------- Accuracy: 0.00% (0/0) ====================================================================== 📊 BENCHMARK SUMMARY ====================================================================== coco_detection: mean_iou: 0.1759 accuracy: 0.0272 num_samples: 100 counting: exact_accuracy: 0.0000 within_one_accuracy: 0.0000 mae: 0.0000 total: 0 vqa: accuracy: 0.0000 correct: 0 total: 0 💾 Results saved to: checkpoints/oculus_detection_v2/final/benchmark_results.json