File size: 5,243 Bytes
67f71c2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
923d2e5
 
 
 
 
67f71c2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4189d2a
67f71c2
 
 
 
 
 
 
 
 
 
 
4189d2a
67f71c2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
"""
Shader environment implementation.

Reference-image-conditioned shader generation with SSIM reward.
"""

import random
from base64 import b64encode
from io import BytesIO
from uuid import uuid4

from openenv.core.env_server.interfaces import Environment
from openenv.core.env_server.types import State
from PIL import Image

try:
    from ..models import ShaderAction, ShaderObservation
    from ..tasks import Task, load as load_tasks, CURATED, CURATED_BY_NAME
    from ..reward import ssim
    from ..harness import render as harness_render
except ImportError:
    from models import ShaderAction, ShaderObservation
    from tasks import Task, load as load_tasks, CURATED, CURATED_BY_NAME
    from reward import ssim
    from harness import render as harness_render


class ShaderEnvironment(Environment):
    """
    OpenEnv environment for shader generation against a reference image.

    Each episode picks a task (shader with known ground-truth code), renders
    the reference frame, then challenges the agent to reproduce it via GLSL.
    """

    SUPPORTS_CONCURRENT_SESSIONS: bool = True
    _tasks: list[Task] | None = None

    def __init__(self, budget: int = 10, seed: int | None = None):
        self._budget = budget
        self._rng = random.Random(seed)
        if ShaderEnvironment._tasks is None:
            try:
                ShaderEnvironment._tasks = load_tasks()
            except FileNotFoundError:
                # Corpus not available — use curated tasks only
                ShaderEnvironment._tasks = list(CURATED)
        self._state = State(episode_id=None, step_count=0)

        # Episode state
        self._task: Task | None = None
        self._ref: bytes | None = None
        self._remaining: int = 0

    def reset(self, seed: int | None = None, episode_id: str | None = None,
              task: str | None = None, **kwargs) -> ShaderObservation:
        """Start a new episode.

        Args:
            seed: RNG seed for reproducible task selection.
            episode_id: Optional episode identifier.
            task: If given, select a curated task by name (e.g. "gradient",
                  "rings", "mandelbrot"). Otherwise picks randomly from the
                  full corpus.
        """
        if seed is not None:
            self._rng = random.Random(seed)

        self._state = State(
            episode_id=episode_id or str(uuid4())[:8],
            step_count=0,
        )
        self._remaining = self._budget

        # Select task(s) to try
        if task and task in CURATED_BY_NAME:
            candidates = [CURATED_BY_NAME[task]]
        else:
            indices = list(range(len(self._tasks)))
            self._rng.shuffle(indices)
            candidates = [self._tasks[i] for i in indices]

        for t in candidates:
            result = harness_render(
                t.code,
                resolution=t.resolution,
                time=t.time,
            )
            if result.compiled and result.rendered and result.frame:
                self._task = t
                self._ref = result.frame
                return ShaderObservation(
                    task=t.name,
                    remaining=self._remaining,
                    reference_png=self._encode(
                        result.frame, result.width, result.height,
                    ),
                    done=False,
                    reward=None,
                )

        raise RuntimeError("no task rendered successfully")

    def step(self, action: ShaderAction) -> ShaderObservation:
        """Render agent's GLSL, compute SSIM vs reference, return observation."""
        if self._task is None:
            raise RuntimeError("call reset() before step()")

        self._state.step_count += 1
        self._remaining -= 1

        result = harness_render(
            action.code,
            resolution=self._task.resolution,
            time=self._task.time,
        )

        # Compute reward in (0, 1) exclusive — validator rejects 0.0 and 1.0
        if result.compiled and result.rendered and result.frame:
            score = ssim(
                self._ref, result.frame,
                result.width, result.height,
            )
            png = self._encode(
                result.frame, result.width, result.height,
            )
        else:
            score = 0.0
            png = ""
        score = min(max(score, 0.01), 0.99)

        done = self._remaining <= 0 or score >= 0.99

        return ShaderObservation(
            task=self._task.name,
            remaining=self._remaining,
            reference_png="",
            compiled=result.compiled,
            rendered=result.rendered,
            errors=result.errors,
            agent_png=png,
            ssim=score,
            done=done,
            reward=score,
        )

    @property
    def state(self) -> State:
        return self._state

    @staticmethod
    def _encode(rgba: bytes, width: int, height: int) -> str:
        """Convert raw RGBA bytes to a base64-encoded PNG string."""
        img = Image.frombytes("RGBA", (width, height), rgba)
        buf = BytesIO()
        img.save(buf, format="PNG")
        return b64encode(buf.getvalue()).decode("ascii")