forked from NervanaSystems/aeon
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathChangeLog
133 lines (107 loc) · 4.44 KB
/
ChangeLog
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
# ChangeLog
## v1.3.1 (2018-01-09)
* Code changes:
* bugfix: of cmake showing error when it could not find Sphinx
* bugfix: cmake DEBUG flag has proper case sensitivity now (Debug instead of DEBUG)
* copyright notice headers updated
* added diagnostic endpoint to aeon service
* (optional) Python plugins performing transformations added
* can be called after aeon built-in transformations.
* created directory examples/plugins with example usage
* created "plugins" directory with example plugins
* Documentation changes:
* doc update with libssl-dev requirement
* support for Ubuntu 14.04 LTS ended, recommended update to 16.04 LTS
* aeon service doc update (openfabrics installation, health endpoint)
* docs on Python plugins:
* documentation doc/source/augmentation_plugins.rst
* New dependencies:
* For Python plugins:
* yaml
* OpenCV version matching Python version
* Sphinx and breathe no longer required for source building (only for docs)
* API changes:
* For Python plugins:
* "plugin_filename" field added under "augmentations"
* "plugin_params" filed added under "augmentations"
* For aeon service:
* health endpoint added
## v1.3.0 (2017-12-13)
* Code changes:
* Distributed dataloading using HTTP or RDMA transfer
* Re-enabling of NDS support
* cmake files improvements
* API changes:
* session_id field added
* New dependencies:
* boost-filesystem
* boost-system
* For distributed:
* cpprestsdk
* libfabric (optional)
## v1.2.0 (2017-11-24)
* Batch transpose implemented in aeon and configurable with `batch_major` field
## v1.1.0 (2017-11-13)
* Working deterministic mode
* Shared thread pool
* Refactoring of box classes
## v1.0.0 (2017-09-19)
* Internal Changes
* No longer rely upon a python backend for initiating transfers to device (all prepared data now remains on host)
* Use of async_manager class to manage all double-buffered workflow
* Removal of support for post-processing of a decoded and loaded batch (was only used for packing transcripts)
* Performance optimizations for batch processing pipeline
* Enhancements
* Modular providers supported by mixing and matching etl components
* Direct c++ interface to loader as an iterator
* Python loader implemented via python c-api
* Unicode support for char_map
* support for SSD topology
## v0.2.7 (2017-03-02)
* Bug fix for clean termination
## v0.2.6 (2016-11-16)
* Bug fix for clean termination
## v0.2.5 (2016-11-15)
* General code cleanup
* Add prefetch to double-buffer reads for NDS and file block reading
* Add class to manage python gil state
* Fix compilation on centos 7
## v0.2.4 (2016-11-02)
* Bug fix for subset sampling of manifest files when block size is small
* In general, more robust handling of subset sampling is now done over entire manifest
* Associated unit tests
* Bug fix for format handling in sox when running in
* Prevent multiple processes from writing to same cache at the same time
* Use crc to compute cache hash based on manifest content instead of modification time
* Associated unit tests
* Add configurable unkown value for character mapping ETL pipeline
* Associated unit tests
## v0.2.3 (2016-10-18)
* Add noise_root path option for relative noise index files
* Bug fix for filterbank shaping
* Bug fix for stereo image provider creation
## v0.2.2 (2016-10-13)
* Add manifest_root path option for cvs manifest files
* Bug fixes for python 3 compatibility
* Retain reference count on python backend object
* Refactor protobackends and tests
## v0.2.1 (2016-10-07)
* Localization improvements
* Bug fix: Handle >64 ground truth bounding boxes
* Image improvements
* Add 'hue' argument to image augmentation where 'hue' can be set from 0..360 degrees
* Stereo provider improvements
* stereo_image,blob provides direct passthrough of arbitrary binary blob
## v0.2.0 (2016-09-23):
* Improved documentation
* Audio improvements
* Sox based extraction to handle more formats natively
* Time domain output
* Bug Fix: normalization for spectral outputs to ignore padded regions
* Image improvements
* Unified image handling for variable sized images and fixed output sizes
* Bug fix: correct photometric augmentation
* Additional provider types
* Localization provider for faster-rcnn style networks
* Stereo image provider
* Bug fix: Better handling of subset percentage for caching loader