# dipin-adhikari/see2seek architecture

> Architecture diagram of the GitHub repository dipin-adhikari/see2seek, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-10-03). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/dipin-adhikari/see2seek
- Repository: https://github.com/dipin-adhikari/see2seek

## Overview

See2Seek trains and evaluates a recurrent agent for zero-shot RoboTHOR navigation: visual observations and image or text goals feed a multimodal policy, which selects actions in simulated episodes; evaluation reports navigation performance. Supporting capabilities include episodic spatial memory, goal-image/embedding preparation, and training-curve plotting. The graph uses README statements and SOURCE INDEX references for relationships; unsampled implementations are represented only where the tree or documentation supports them.

Main entry points: [`see2seek/trainers/ppo_trainer.py`](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/trainers/ppo_trainer.py) (PPO trainer), [`see2seek/evaluation/evaluator.py`](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/evaluation/evaluator.py) (Evaluator), [`see2seek/models/encoders/clip_encoder.py`](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/models/encoders/clip_encoder.py) (CLIP encoder), [`see2seek/utils/generate_data.py`](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/utils/generate_data.py) (Goal preparation)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_data["Goal data"]
  node_goalprep["Goal preparation<br/>[generate_data.py]"]
  node_angleaug["Angle augmentation"]
  node_goalcache[("Goal cache<br/>[goal_cache.py]")]
end

subgraph group_runtime["Navigation runtime"]
  node_config["Configuration<br/>[config.py]"]
  node_environment["RoboTHOR environment<br/>[robothor_env.py]"]
  node_vecenv["Vector environments<br/>[vec_env.py]"]
end

subgraph group_learning["Policy learning"]
  node_dino["DINOv2 encoder<br/>[dino_encoder.py]"]
  node_clip["CLIP encoder<br/>[clip_encoder.py]"]
  node_policy["GRU policy<br/>[gru_policy.py]"]
  node_memory["Episodic memory<br/>[gru_policy.py]"]
  node_rollouts[("Rollout buffer<br/>[rollout_buffer.py]")]
  node_trainer["PPO trainer<br/>[ppo_trainer.py]"]
end

subgraph group_assessment["Evaluation"]
  node_metrics["Navigation metrics<br/>[metrics.py]"]
  node_evaluator["Evaluator<br/>[evaluator.py]"]
  node_trainingplots["Training plots<br/>[plot_training.py]"]
end

node_user(("Experimenter"))
node_simulator["AI2-THOR"]

node_user -->|"prepares goals"| node_goalprep
node_goalprep -->|"encodes goals"| node_clip
node_goalprep -->|"uses cache"| node_goalcache
node_user -->|"augments views"| node_angleaug
node_angleaug -->|"encodes views"| node_clip
node_user -->|"trains agent"| node_trainer
node_trainer -->|"loads settings"| node_config
node_trainer -->|"collects rollouts"| node_vecenv
node_trainer -->|"encodes observations"| node_dino
node_trainer -->|"encodes goals"| node_clip
node_trainer -->|"optimizes policy"| node_policy
node_trainer -->|"stores experience"| node_rollouts
node_policy -->|"uses memory"| node_memory
node_vecenv -.->|"runs episodes"| node_environment
node_environment -->|"controls scenes"| node_simulator
node_user -->|"evaluates checkpoint"| node_evaluator
node_evaluator -->|"loads settings"| node_config
node_evaluator -->|"runs evaluations"| node_vecenv
node_evaluator -->|"uses episodes"| node_environment
node_evaluator -->|"encodes observations"| node_dino
node_evaluator -->|"encodes goals"| node_clip
node_evaluator -->|"runs policy"| node_policy
node_user -->|"plots training"| node_trainingplots
node_trainingplots -->|"reads settings"| node_config

click node_goalprep "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/utils/generate_data.py"
click node_angleaug "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/utils/augment_goal_angles.py"
click node_goalcache "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/utils/goal_cache.py"
click node_config "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/utils/config.py"
click node_environment "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/envs/robothor_env.py"
click node_vecenv "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/envs/vec_env.py"
click node_dino "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/models/encoders/dino_encoder.py"
click node_clip "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/models/encoders/clip_encoder.py"
click node_policy "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/agents/gru_policy.py"
click node_memory "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/agents/gru_policy.py"
click node_rollouts "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/buffers/rollout_buffer.py"
click node_trainer "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/trainers/ppo_trainer.py"
click node_metrics "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/evaluation/metrics.py"
click node_evaluator "https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/evaluation/evaluator.py"
click node_trainingplots "https://github.com/dipin-adhikari/see2seek/blob/main/scripts/plot_training.py"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_goalprep,node_angleaug,node_goalcache toneBlue
class node_config,node_environment,node_vecenv toneAmber
class node_dino,node_clip,node_policy,node_memory,node_rollouts,node_trainer toneMint
class node_metrics,node_evaluator,node_trainingplots toneRose
class node_user,node_simulator toneIndigo
```

## Components

### Goal data

- [Goal preparation](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/utils/generate_data.py): `see2seek/utils/generate_data.py`
- [Angle augmentation](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/utils/augment_goal_angles.py): `see2seek/utils/augment_goal_angles.py`
- [Goal cache](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/utils/goal_cache.py): `see2seek/utils/goal_cache.py`

### Navigation runtime

- [Configuration](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/utils/config.py): `see2seek/utils/config.py`
- [RoboTHOR environment](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/envs/robothor_env.py): `see2seek/envs/robothor_env.py`
- [Vector environments](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/envs/vec_env.py): `see2seek/envs/vec_env.py`

### Policy learning

- [DINOv2 encoder](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/models/encoders/dino_encoder.py): `see2seek/models/encoders/dino_encoder.py`
- [CLIP encoder](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/models/encoders/clip_encoder.py): `see2seek/models/encoders/clip_encoder.py`
- [GRU policy](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/agents/gru_policy.py): `see2seek/agents/gru_policy.py`
- [Episodic memory](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/agents/gru_policy.py): `see2seek/agents/gru_policy.py`
- [Rollout buffer](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/buffers/rollout_buffer.py): `see2seek/buffers/rollout_buffer.py`
- [PPO trainer](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/trainers/ppo_trainer.py): `see2seek/trainers/ppo_trainer.py`

### Evaluation

- [Navigation metrics](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/evaluation/metrics.py): `see2seek/evaluation/metrics.py`
- [Evaluator](https://github.com/dipin-adhikari/see2seek/blob/main/see2seek/evaluation/evaluator.py): `see2seek/evaluation/evaluator.py`
- [Training plots](https://github.com/dipin-adhikari/see2seek/blob/main/scripts/plot_training.py): `scripts/plot_training.py`

### Other components

- **Experimenter**
- **AI2-THOR**

## Connections

- Experimenter → Goal preparation: prepares goals (evidence: `README.md`)
- Goal preparation → CLIP encoder: encodes goals (evidence: `see2seek/utils/generate_data.py`)
- Goal preparation → Goal cache: uses cache (evidence: `see2seek/utils/generate_data.py`)
- Experimenter → Angle augmentation: augments views (evidence: `README.md`)
- Angle augmentation → CLIP encoder: encodes views (evidence: `see2seek/utils/augment_goal_angles.py`)
- Experimenter → PPO trainer: trains agent (evidence: `README.md`)
- PPO trainer → Configuration: loads settings (evidence: `see2seek/trainers/ppo_trainer.py`)
- PPO trainer → Vector environments: collects rollouts (evidence: `see2seek/trainers/ppo_trainer.py`)
- PPO trainer → DINOv2 encoder: encodes observations (evidence: `see2seek/trainers/ppo_trainer.py`)
- PPO trainer → CLIP encoder: encodes goals (evidence: `see2seek/trainers/ppo_trainer.py`)
- PPO trainer → GRU policy: optimizes policy (evidence: `see2seek/trainers/ppo_trainer.py`)
- PPO trainer → Rollout buffer: stores experience (evidence: `see2seek/trainers/ppo_trainer.py`)
- GRU policy → Episodic memory: uses memory (evidence: `README.md`)
- Vector environments → RoboTHOR environment: runs episodes
- RoboTHOR environment → AI2-THOR: controls scenes (evidence: `see2seek/envs/robothor_env.py`)
- Experimenter → Evaluator: evaluates checkpoint (evidence: `README.md`)
- Evaluator → Configuration: loads settings (evidence: `see2seek/evaluation/evaluator.py`)
- Evaluator → Vector environments: runs evaluations (evidence: `see2seek/evaluation/evaluator.py`)
- Evaluator → RoboTHOR environment: uses episodes (evidence: `see2seek/evaluation/evaluator.py`)
- Evaluator → DINOv2 encoder: encodes observations (evidence: `see2seek/evaluation/evaluator.py`)
- Evaluator → CLIP encoder: encodes goals (evidence: `see2seek/evaluation/evaluator.py`)
- Evaluator → GRU policy: runs policy (evidence: `see2seek/evaluation/evaluator.py`)
- Experimenter → Training plots: plots training (evidence: `README.md`)
- Training plots → Configuration: reads settings (evidence: `scripts/plot_training.py`)
