kobiakor15 commited on
Commit
4145f82
·
verified ·
1 Parent(s): bdb0102

Upload docs/TRAINING_ROADMAP.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. docs/TRAINING_ROADMAP.md +17 -0
docs/TRAINING_ROADMAP.md ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # 🚀 Oculus V3: Future Training Roadmap
2
+
3
+ COCO (Current) = 80 common classes. Good baseline, but limited for real-world niche tasks.
4
+
5
+ ## Option A: Universal Detection (The "Scanner")
6
+ **Target**: Detect 1200+ specific objects.
7
+ - **Dataset**: **LVIS** or **Objects365**.
8
+ - **Result**: Recognizes "stapler", "doorknob", "mango" instead of just generic classes.
9
+
10
+ ## Option B: Visual Reasoning (The "Thinker")
11
+ **Target**: Better VQA and complex instruction following.
12
+ - **Dataset**: **LLaVA-Instruct** or **VizWiz**.
13
+ - **Why**: Teaches the model to "explain why the car is parked" rather than just finding the car.
14
+ - **Result**: A smarter chatbot-like VLM.
15
+
16
+ ## Recommendation
17
+ Since Oceanir is a VLM platform, **Option B (Instruction Tuning)** is the highest value next step. It improves the model's IO (Intelligence Output) significantly more than just adding more bounding boxes.