File size: 8,098 Bytes
744eb4e |
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 |
from collections import OrderedDict, defaultdict
import transformers
from pointllm import conversation as conversation_lib
from dataclasses import dataclass
from typing import Optional, Dict, Sequence
import torch
import numpy as np
import os
IGNORE_INDEX = -100
# * Sample Usage:
# * from utils import LRUCache
# * cache = LRUCache(capacity, max_access_count)
# if self.cache is None:
# info_data = self.multiview_scannet[info_index]
# else:
# info_data = self.cache.get(info_index)
# if info_data is None or self.cache.get_access_count(info_index) >= self.cache.max_access_count:
# # If not in cache, or accessed max_access_count times, load it and put it in cache
# info_data = self.multiview_scannet[info_index]
# self.cache.put(info_index, info_data)
# self.cache.reset_access_count(info_index)
class LRUCache:
def __init__(self, capacity, max_access_count):
self.cache = OrderedDict()
self.access_count = defaultdict(int)
self.capacity = capacity
self.max_access_count = max_access_count
def get(self, key):
if key not in self.cache:
return None
value = self.cache.pop(key)
self.cache[key] = value # Put key as the newest one
self.access_count[key] += 1
return value
def put(self, key, value):
if key in self.cache: # Update the value and put it as newest
self.cache.pop(key)
elif len(self.cache) == self.capacity: # If cache is full
oldest_key = next(iter(self.cache))
self.cache.popitem(last=False) # Remove oldest item
del self.access_count[oldest_key] # Remove the corresponding access count
self.cache[key] = value
self.access_count[key] = 1
def get_access_count(self, key):
return self.access_count.get(key, 0)
def reset_access_count(self, key):
self.access_count[key] = 0
def preprocess_v1(
sources,
tokenizer: transformers.PreTrainedTokenizer,
) -> Dict:
conv = conversation_lib.default_conversation.copy()
roles = {"human": conv.roles[0], "gpt": conv.roles[1]}
# Apply prompt templates
conversations = []
for i, source in enumerate(sources):
if roles[source[0]["from"]] != conv.roles[0]:
# Skip the first one if it is not from human
source = source[1:]
conv.messages = []
for j, sentence in enumerate(source):
role = roles[sentence["from"]]
assert role == conv.roles[j % 2], f"{i}"
conv.append_message(role, sentence["value"])
conversations.append(conv.get_prompt())
# Tokenize conversations
input_ids = tokenizer(
conversations,
return_tensors="pt",
padding="longest",
max_length=tokenizer.model_max_length,
truncation=True,
).input_ids
targets = input_ids.clone()
assert conv.sep_style == conversation_lib.SeparatorStyle.TWO
# Mask targets
sep = conv.sep + conv.roles[1] + ": "
for conversation, target in zip(conversations, targets):
total_len = int(target.ne(tokenizer.pad_token_id).sum())
rounds = conversation.split(conv.sep2)
cur_len = 1
target[:cur_len] = IGNORE_INDEX
for i, rou in enumerate(rounds):
if rou == "":
break
parts = rou.split(sep)
if len(parts) != 2: # * can handle padded tokens
break
parts[0] += sep
round_len = len(tokenizer(rou).input_ids)
instruction_len = len(tokenizer(parts[0]).input_ids) - 2
target[cur_len : cur_len + instruction_len] = IGNORE_INDEX
cur_len += round_len
target[cur_len:] = IGNORE_INDEX # * this is necessary for padded tokens
if cur_len < tokenizer.model_max_length:
if cur_len != total_len: # * unk tokens in the dialogue will cause this.
target[:] = IGNORE_INDEX
print(
f"WARNING: tokenization mismatch: {cur_len} vs. {total_len}."
f" (ignored)"
)
return dict(
input_ids=input_ids,
labels=targets,
)
def preprocess_multimodal_point_cloud(
sources: Sequence[str],
point_backbone_config: dict,
point_indicator: str = "<point>",
) -> Dict:
point_token_len = point_backbone_config['point_token_len']
default_point_patch_token = point_backbone_config['default_point_patch_token']
for source in sources:
for sentence in source:
replace_token = default_point_patch_token * point_token_len
if point_backbone_config['mm_use_point_start_end']:
replace_token = point_backbone_config['default_point_start_token']+ replace_token + point_backbone_config['default_point_end_token']
sentence["value"] = sentence["value"].replace(point_indicator, replace_token)
return sources
def pc_norm(pc):
""" pc: NxC, return NxC """
xyz = pc[:, :3]
other_feature = pc[:, 3:]
centroid = np.mean(xyz, axis=0)
xyz = xyz - centroid
m = np.max(np.sqrt(np.sum(xyz ** 2, axis=1)))
xyz = xyz / m
pc = np.concatenate((xyz, other_feature), axis=1)
return pc
def load_objaverse_point_cloud(data_path, object_id, pointnum=8192, use_color=False):
filename = f"{object_id}_{pointnum}.npy"
point_cloud = np.load(os.path.join(data_path, filename))
# * normalize
point_cloud = pc_norm(point_cloud)
if not use_color:
point_cloud = point_cloud[:, :3]
return point_cloud
@dataclass
class DataCollatorForPointTextDataset(object):
"""Collate examples for mixed dataset with text and point cloud data."""
tokenizer: transformers.PreTrainedTokenizer
def __call__(self, instances: Sequence[Dict]) -> Dict[str, torch.Tensor]:
input_ids, labels = tuple([instance[key] for instance in instances]
for key in ("input_ids", "labels"))
input_ids = torch.nn.utils.rnn.pad_sequence(
input_ids,
batch_first=True,
padding_value=self.tokenizer.pad_token_id)
labels = torch.nn.utils.rnn.pad_sequence(labels,
batch_first=True,
padding_value=IGNORE_INDEX)
batch = dict(
input_ids=input_ids,
labels=labels,
attention_mask=input_ids.ne(self.tokenizer.pad_token_id),
)
if 'point_clouds' in instances[0]:
point_clouds = [instance['point_clouds'] for instance in instances]
if all(x is not None and x.shape == point_clouds[0].shape for x in point_clouds): # * point_clouds have different shapes
batch['point_clouds'] = torch.stack(point_clouds)
else:
batch['point_clouds'] = point_clouds # * return as lists
return batch
def farthest_point_sample(point, npoint):
"""
Input:
xyz: pointcloud data, [N, D]
npoint: number of samples
Return:
centroids: sampled pointcloud index, [npoint, D]
"""
N, D = point.shape
xyz = point[:,:3]
centroids = np.zeros((npoint,))
distance = np.ones((N,)) * 1e10
farthest = np.random.randint(0, N)
for i in range(npoint):
centroids[i] = farthest
centroid = xyz[farthest, :]
dist = np.sum((xyz - centroid) ** 2, -1)
mask = dist < distance
distance[mask] = dist[mask]
farthest = np.argmax(distance, -1)
point = point[centroids.astype(np.int32)]
return point
def pc_normalize(pc):
"""
pc: Nx3 array
This functions normalizes a point cloud to fit within a unit sphere.
It first calculates the centroid of the point cloud and then subtracts
it from all points before scaling all points to fit within a unit sphere.
"""
centroid = np.mean(pc, axis=0)
pc = pc - centroid
m = np.max(np.sqrt(np.sum(pc**2, axis=1)))
pc = pc / m
return pc |