Upload folder using huggingface_hub

README.md

@@ -0,0 +1,24 @@
+# LOKI — HDINO-T Open-Vocabulary Detector
+
+**Model**: HDINO-T (Swin-T backbone + deformable transformer)
+**Paper**: [HDINO: A Concise and Efficient Open-Vocabulary Detector](https://hf.co/papers/2603.02924)
+**Training**: Fine-tuned on COCO 2017 (118K images, 80 classes)
+**Best val_loss**: 2.36 (epoch 2)
+
+## Files
+- `hdino_t_best.pth` — Full checkpoint (model + optimizer + scheduler)
+- `hdino_t_best.safetensors` — Model weights only (safetensors format)
+- `hdino_t_best.pt` — Model bundle (weights + config + text embeddings)
+
+## Training Config
+- LR: 0.0001 (cosine decay)
+- Batch size: 6
+- Precision: bf16
+- Optimizer: AdamW
+- CLIP text encoder: frozen (openai/clip-vit-base-patch32)
+
+## Usage
+Requires HDINO repo: https://github.com/HaoZ416/HDINO
+
+## Part of ANIMA Wave-6
+Module: project_loki | Tier 2 Perception

hdino_t_best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7808478b6c8b11ecced55054518569df28eb7ca65462dc08599ff44e83b2c53
+size 452461847

hdino_t_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9bc726c969c9c039da21c516c547ceaa7397507c31c0b8968099b022bd8e7423
+size 830141040

hdino_t_best.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7275922af385f8e5b535bec18a91f2b6d562c727645389804514e66970cb5342
+size 446387272