File size: 14,173 Bytes
fb11af9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
from typing import Dict

import numpy as np
import torch
import math
import einops
import torch.nn as nn
import torch.nn.functional as F
from torch import Tensor


IMAGE_KEYS = (
    "base_0_rgb",
    "left_wrist_0_rgb",
    "right_wrist_0_rgb",
)


def dict_apply(func, d):
    """
    Apply a function to all values in a dictionary recursively.
    If the value is a dictionary, it will apply the function to its values.
    """
    for key, value in d.items():
        if isinstance(value, dict):
            dict_apply(func, value)
        else:
            d[key] = func(value)
    return d

class Normalizer:
    def __init__(
        self,
        norm_stats: Dict[str, Dict[str, np.ndarray]],
        from_file: bool=False,
        data_type: str=None,
        norm_type: Dict[str, str] | None = None,
    ):
        if from_file:
            if data_type == 'libero':
                norm_stats['state']['mean'] = np.array(norm_stats['state']['mean'][:8])
                norm_stats['state']['std'] = np.array(norm_stats['state']['std'][:8])
                norm_stats['actions']['mean'] = np.array(norm_stats['actions']['mean'][:7])
                norm_stats['actions']['std'] = np.array(norm_stats['actions']['std'][:7])
            elif data_type == 'robotwin':
                norm_stats['observation.state'], norm_stats['action']  = {}, {}
                norm_stats['observation.state']['q01'] = np.array(norm_stats['observation.state.arm.position']['q01'][:6] + norm_stats['observation.state.effector.position']['q01'][:1] + norm_stats['observation.state.arm.position']['q01'][6:] + norm_stats['observation.state.effector.position']['q01'][1:])
                norm_stats['observation.state']['q99'] = np.array(norm_stats['observation.state.arm.position']['q99'][:6] + norm_stats['observation.state.effector.position']['q99'][:1] + norm_stats['observation.state.arm.position']['q99'][6:] + norm_stats['observation.state.effector.position']['q99'][1:])
                norm_stats['action']['q01'] = np.array(norm_stats['action.arm.position']['q01'][:6] + norm_stats['action.effector.position']['q01'][:1] + norm_stats['action.arm.position']['q01'][6:] + norm_stats['action.effector.position']['q01'][1:])
                norm_stats['action']['q99'] = np.array(norm_stats['action.arm.position']['q99'][:6] + norm_stats['action.effector.position']['q99'][:1] + norm_stats['action.arm.position']['q99'][6:] + norm_stats['action.effector.position']['q99'][1:])
            elif data_type == 'robotwin_rep':
                norm_stats['observation.state'], norm_stats['action']  = {}, {}
                norm_stats['observation.state']['q01'] = np.array(norm_stats['observation.state.arm.position']['q01'] + norm_stats['observation.state.effector.position']['q01'])
                norm_stats['observation.state']['q99'] = np.array(norm_stats['observation.state.arm.position']['q99'] + norm_stats['observation.state.effector.position']['q99'])
                norm_stats['action']['q01'] = np.array(norm_stats['action.arm.position']['q01'][:6] + norm_stats['action.effector.position']['q01'][:1] + norm_stats['action.arm.position']['q01'][6:] + norm_stats['action.effector.position']['q01'][1:])
                norm_stats['action']['q99'] = np.array(norm_stats['action.arm.position']['q99'][:6] + norm_stats['action.effector.position']['q99'][:1] + norm_stats['action.arm.position']['q99'][6:] + norm_stats['action.effector.position']['q99'][1:])
            elif data_type == 'customized':
                for key in norm_stats:
                    if isinstance(norm_stats[key], dict):
                        for sub_key in norm_stats[key]:
                            norm_stats[key][sub_key] = np.array(norm_stats[key][sub_key])
            self.norm_stats = norm_stats
        else:
            self.norm_stats = dict_apply(lambda x: x.astype(np.float32), norm_stats)
        self.norm_type = norm_type or {}
        self.from_file = from_file

    def normalize(self, data: Dict[str, np.ndarray]) -> Dict[str, torch.Tensor]:
        normalized_data = {}
        for key, value in data.items():
            if key in self.norm_stats:
                norm_type = self.norm_type.get(key, "identity")
                if norm_type == "meanstd":
                    mean = self.norm_stats[key]["mean"]
                    std = self.norm_stats[key]["std"]
                    normalized_value = (value - mean) / (std + 1e-6)
                elif norm_type == "bounds_99_woclip":
                    low = self.norm_stats[key]["q01"]
                    high = self.norm_stats[key]["q99"]
                    normalized_value = (value  - low) / (high - low + 1e-6) * 2.0 - 1.0
                elif norm_type == "std":
                    std = self.norm_stats[key]["std"]
                    normalized_value = value / (std + 1e-6)
                elif norm_type == "minmax":
                    min_val = self.norm_stats[key]["min"]
                    max_val = self.norm_stats[key]["max"]
                    normalized_value = (value - min_val) / (
                        max_val - min_val + 1e-6
                    ) * 2 - 1
                elif norm_type == "identity":
                    normalized_value = value
                else:
                    raise ValueError(
                        f"Unknown normalization type: {norm_type}. Supported types are 'meanstd', 'minmax', and 'identity'."
                    )
                normalized_data[key] = normalized_value
            else:
                # If the key is not in norm_stats, we assume no normalization is needed
                normalized_data[key] = value
        return normalized_data

    def unnormalize(self, data: Dict[str, np.ndarray]) -> Dict[str, np.ndarray]:
        """
        Unnormalize the given data using stored normalization statistics.

        Args:
            data (Dict[str, np.ndarray]): Dictionary of normalized arrays to unnormalize.

        Returns:
            Dict[str, np.ndarray]: Dictionary of unnormalized arrays.
        """
        unnormalized_data = {}
        for key, value in data.items():
            if key in self.norm_stats:
                norm_type = self.norm_type.get(key, "identity")
                stats = self.norm_stats[key]
                if norm_type == "meanstd":
                    mean = stats["mean"]
                    std = stats["std"]
                    unnormalized_value = value * (std + 1e-6) + mean
                elif norm_type == "bounds_98" or norm_type == 'bounds_98_woclip':
                    low = self.norm_stats[key]["q02"]
                    high = self.norm_stats[key]["q98"]
                    unnormalized_value = ((value + 1.0) / 2.0) * (high - low + 1e-6) + low
                elif norm_type == "bounds_99" or norm_type == "bounds_99_woclip":
                    low = self.norm_stats[key]["q01"]
                    high = self.norm_stats[key]["q99"]
                    unnormalized_value = ((value + 1.0) / 2.0) * (high - low + 1e-6) + low
                elif norm_type == "std":
                    std = stats["std"]
                    unnormalized_value = value * (std + 1e-6)
                elif norm_type == "minmax":
                    min_val = stats["min"]
                    max_val = stats["max"]
                    # Reverse: (x + 1)/2 * (max-min+eps) + min
                    unnormalized_value = (value + 1) / 2.0 * (max_val - min_val + 1e-6) + min_val
                elif norm_type == "identity":
                    unnormalized_value = value
                else:
                    raise ValueError(
                        f"Unknown normalization type: {norm_type}. Supported types are 'meanstd', 'minmax', and 'identity'."
                    )
                unnormalized_data[key] = unnormalized_value
            else:
                # If no normalization was applied, return as-is
                unnormalized_data[key] = value
        return unnormalized_data

def resize_with_pad_item(img, width, height, pad_value=-1):
    # assume no-op when width height fits already
    if img.ndim != 3:
        raise ValueError(f"(c,h,w) expected, but {img.shape}")

    cur_height, cur_width = img.shape[1:]

    ratio = max(cur_width / width, cur_height / height)
    resized_height = int(cur_height / ratio)
    resized_width = int(cur_width / ratio)
    resized_img = F.interpolate(
        img.unsqueeze(0), size=(resized_height, resized_width), mode="bilinear", align_corners=False
    ).squeeze(0)

    pad_height = max(0, int(height - resized_height))
    pad_width = max(0, int(width - resized_width))

    # pad on left and top of image
    padded_img = F.pad(resized_img, (pad_width, 0, pad_height, 0), value=pad_value)
    return padded_img

def prepare_images(config, image_processor, observation: dict[str, Tensor], use_depth_align=False):
    """Normalize, resize, and pad images and stack them into a tensor.

    Args:
        observation (dict[str, Tensor])

    Returns:
        images (torch.Tensor): (*b, n, c, h, w) images in range [-1.0, 1.0]
        img_masks (torch.Tensor): (*b, n) masks for images, True if image is present, False if missing
    """
    dtype = observation["state"].dtype
    images, img_masks = [], []
    if use_depth_align:
        pil_images = []

    for key in IMAGE_KEYS:
        if key in observation["image"]:
            # resize, pad, and normalize
            img = observation["image"][key]
            assert img.ndim == 3, f"Expected 3D image, got {img.shape}"
            pil_img = img.cpu().numpy()
            if image_processor is None:
                img = img.to(dtype) / 127.5 - 1.0 # to [-1, 1]
                img = resize_with_pad_item(
                    img, *config.resize_imgs_with_padding, pad_value=-1.0
                )
            else:
                img = resize_with_pad_item(
                    img, *config.resize_imgs_with_padding, pad_value=0
                )
                img = image_processor(img)['pixel_values']
            images.append(img)
            img_masks.append(True)
            if use_depth_align:
                pil_images.append(pil_img)
        else:
            # zero padding
            if image_processor is None:
                img = torch.full_like(img, fill_value=-1.0)
                if use_depth_align:
                    pil_img = torch.full_like(pil_img, fill_value=-1.0)
            else:
                img = np.zeros_like(img)
                if use_depth_align:
                    pil_img = np.zeros_like(pil_img)
            images.append(img)
            if use_depth_align:
                pil_images.append(pil_img)
            img_masks.append(False)
    if isinstance(images[0], torch.Tensor):
        images = torch.stack(images, dim=0)  # (n, c, h, w)
    elif isinstance(images[0], np.ndarray):
        images = torch.from_numpy(np.stack(images, axis=0))  # (n, c, h, w)
    img_masks = torch.tensor(img_masks, dtype=torch.bool)  # (*n)

    if use_depth_align:
        pil_images = torch.from_numpy(np.stack(pil_images, axis=0))  # (n, c, h, w)
    else:
        pil_images = []

    return images, img_masks, pil_images

def prepare_state(config, observation: dict[str, Tensor]):
    """Pad the state to the maximum state dimension.

    Args:
        observation (dict[str, Tensor])

    Returns:
        state (torch.Tensor): (*b, max_state_dim) padded state tensor
    """
    state = observation["state"]
    state = F.pad(state, (0, config.max_state_dim - state.shape[-1]))
    return state

def prepare_action(config, observation: dict[str, Tensor]):
    """Pad the action to the maximum action dimension.

    Args:
        observation (dict[str, Tensor])

    Returns:
        action (torch.Tensor): (*b, n, max_action_dim) padded action tensor
        action_dim (int): the actual dimension of the action before padding
    """
    # ipdb.set_trace()
    action = observation["action"]
    action = F.pad(action, (0, config.max_action_dim - action.shape[-1]))
    return action

def prepare_language(config, language_tokenizer, observation: dict[str, Tensor]):
    """If `prompt` is provided, modify it to PaliGemma format and tokenize it.
    If `lang_tokens` and `lang_masks` are provided, use them directly.

    PaliGemma expects prefix prompts to be formatted as:
    <images> .... <images> <bos> prompt <sep>, where <sep> uses `\\n`.
    So here we format the prompt to start with `<bos>` and end with `\\n`.
    Later, we will concatenate the images and language tokens into a single sequence.

    Args:
        observation (dict[str, Tensor])

    Returns:
        lang_tokens (torch.Tensor): (*b, l) language tokens
        lang_masks (torch.Tensor): (*b, l) masks for language tokens, True if token is present, False if missing
    """
    lang_tokens = observation.get("lang_tokens", None)
    lang_masks = observation.get("lang_masks", None)
    prompt = observation.get("prompt", None)

    # either provide `prompt` or (`lang_tokens`, `lang_masks`)
    if prompt is None and (lang_tokens is None or lang_masks is None):
        raise ValueError(
            "Either 'prompt' or ('lang_tokens', 'lang_masks') must be provided in the observation."
        )

    device = observation["state"].device
    if prompt is not None and (lang_tokens is None or lang_masks is None):
        prompt = [p if p.startswith("<bos>") else f"<bos>{p}" for p in prompt]
        prompt = [p if p.endswith("\n") else f"{p}\n" for p in prompt]
        tokenized_prompt = language_tokenizer.__call__(
            prompt,
            padding="max_length",
            padding_side="right",
            max_length=config.tokenizer_max_length,
            truncation=True,
            return_tensors="pt",
        )
        lang_tokens = tokenized_prompt["input_ids"].to(device=device)
        lang_masks = tokenized_prompt["attention_mask"].to(
            device=device, dtype=torch.bool
        )
    else:
        lang_tokens = observation["lang_tokens"].to(device=device)
        lang_masks = observation["lang_masks"].to(device=device, dtype=torch.bool)

    return lang_tokens.squeeze(0), lang_masks.squeeze(0)