Datumbox offers an innovative Machine Learning platform specialized in Natural Language Processing.
Joined July 2013
- Tweets 7,811
- Following 42
- Followers 626
- Likes 46
2 Photos and videos
Datumbox retweeted
26 Apr 2025
I've open-sourced VernamVeil, an experimental cipher written in pure Python, designed for developers curious about cryptography’s inner workings. It’s only about 200 lines of Python code with no external dependencies other than standard Python libraries: blog.datumbox.com/vernamveil…
1
2
235
TorchVision’s newest API addresses a long-standing request from the community. Read on to find out how it can be used to list and initialize available models & weights by name: bit.ly/3QykfD5
3
18
125
Datumbox retweeted
15 Aug 2022
debuggercafe.com/comparing-p…
New post on DebuggerCafe - Comparing PyTorch ImageNetV1 and ImageNetV2 Weights with Torchvision 0.13
#PyTorch #Torchvision #DeepLearning #TransferLearning
4
4
Datumbox retweeted
10 Aug 2022
We just merged to TorchVision the SwinTransformerV2 & MViTv2 models, along with pre-trained weights:
- swin_v2_t 82.1%
- swin_v2_s 83.7%
- swin_v2_b 84.1%
- mvit_v2_s 80.8%
Many thanks to Ren Pang and Yanghao Li for their work:
github.com/pytorch/vision/pu…
github.com/pytorch/vision/pu…
2
10
Datumbox retweeted
8 Aug 2022
debuggercafe.com/model-libra…
New post on DebuggerCafe - Model Library Updates with PyTorch 1.12. A summary of the major changes that took place with new PyTorch 1.12
#PyTorch #torchvision #DeepLearning
5
6
Datumbox retweeted
7 Aug 2022
Natural Language Processing with Groovy, OpenNLP, CoreNLP, Nlp4j, Datumbox, Smile, Spark NLP, DJL and TensorFlow
blogs.apache.org/groovy/entr…
#groovylang @ApacheSpark @ApacheGroovy @deepjavalibrary @TensorFlow @ApacheOpennlp @stanfordnlp @haifengl @Datumbox @oyahiroki
2
20
39
Datumbox retweeted
29 Jun 2022
TorchVision v0.13 and @PyTorch v1.12 released! 🥳
- New Multi-weight support API with meta-data & preprocessing transforms
- Swin Transformer & EfficientNetV2 models
- New improved pretrained weights for all popular models
- New AugMix, DropBlock & losses
pytorch.org/blog/pytorch-1.1…
31
131
Datumbox retweeted
24 Jun 2022
We have just merged the MViT architecture in TorchVision. This transformer video model achieves an 78.5% accuracy on Kinetics400. github.com/pytorch/vision/pu…
Image variants to be released soon. Many thanks to the @PyTorch Video team for their support!
#ComputerVision #MachineLearning
6
21
Datumbox retweeted
15 Jun 2022
We just merged to @PyTorch's TorchVision an experimental implementation of the Simple Copy-Paste data augmentation that aims to improve the detection models by 1.5 mAP: github.com/pytorch/vision/pu…
Massive thanks to our community contributor @Lezwon #ComputerVision #MachineLearning
1
9
26
Datumbox retweeted
21 May 2022
Here is my latest blog post on @PyTorch's journey to Modernize TorchVision: blog.datumbox.com/the-journe…
#MachineLearning #ComputerVision #DataScience
2
39
121
Datumbox retweeted
27 Apr 2022
The Swin Transformer model was just merged in TorchVision, along with pre-trained weights for swin_t (Acc@1 81.4%). More variants coming soon!
Massive thanks to Hu Ye for fully reproducing the results of the paper: github.com/pytorch/vision/pu… #PyTorch #DataScience #ComputerVision
6
15
Datumbox retweeted
6 Apr 2022
TorchVision refreshed its most popular Detection & Segmentation models to improve their accuracies:
- RetinaNet: 41.5 mAP ( 5.1)
- FasterRCNN: 46.7 mAP ( 9.7)
- MaskRCNN: 47.4 box mAP ( 9.5) / 41.8 mask mAP ( 7.2)
github.com/pytorch/vision/is…
#MachineLearning #ComputerVision #PyTorch
1
5
10
Datumbox retweeted
22 Mar 2022
Say hello to TorchVision's new Multi-weight support API!!
It's fully-BC, enables support of multiple pre-trained weights for the same model variant, bundles together the preprocessing transforms & other crucial meta-data: github.com/pytorch/vision/pu…
#MachineLearning #ComputerVision
5
10
Datumbox retweeted
2 Mar 2022
The EfficientNetV2 architecture was just merged in TorchVision. We offer pretained weights on ImageNet with the following Acc@1:
- Small 84.2%
- Medium 85.1%
- Large 85.8%
github.com/pytorch/vision/pu… #PyTorch #DataScience #ComputerVision
41
188
Datumbox retweeted
18 Feb 2022
We just merged the AugMix auto-augmentation strategy to TorchVision. Many thanks to @TheNormanMu and @hendrycks for their help & reviews: github.com/pytorch/vision/pu…
We also added the Large Scale Jitter transform for Object Detection:
github.com/pytorch/vision/pu…
#PyTorch #DataScience
2
11
Datumbox retweeted
2 Feb 2022
The ConvNeXt architecture has graduated from prototype into main TorchVision! We offer improved pre-trained weights, using our latest training recipe, with the following Acc@1:
- Tiny 82.5%
- Small 83.6%
- Base 84.1%
- Large 84.4%
github.com/pytorch/vision/pu… #PyTorch #ComputerVision
3
12
Datumbox retweeted
24 Jan 2022
debuggercafe.com/brain-mri-c…
New post at DebuggerCafe - PyTorch EfficientNetB0 for Brain MRI Image Classification
#PyTorch #EfficientNet #EfficientNetB0 #TransferLearning #FineTuning #ImageClassification #DeepLearning
1
3
5
Datumbox retweeted
21 Jan 2022
We just merged FCOS, the object detection architecture, along with pre-trained weights in TorchVision: github.com/pytorch/vision/pu…
Massive thanks to Hu Ye, @zhiq_w and Joao Gomes for their work on fully reproducing the results of the paper: arxiv.org/abs/1904.01355
(1/3)
2
2
4
Datumbox retweeted
17 Jan 2022
I just merged to TorchVision a ResNet50 weights update contributed by @tbennun which pushed the accuracy to 80.9%: github.com/pytorch/vision/pu…
This recipe AdamW achieves 82.5% Acc@1 for the newly released ConvNeXt-T: github.com/pytorch/vision/pu…
#PyTorch #ComputerVision #MachineLearning
1
16
69
Datumbox retweeted
22 Dec 2021
TorchVision introduces a new model API which offers multi-weight support. On the article below, we discuss the proposed changes & request for your feedback.
Help us finalize the API by providing your input:
lnkd.in/eSXpFuVU
#ComputerVision #MachineLearning #PyTorch
TorchVision has released a new backwards compatible API for building models with multi-weight support. In this blog post, we'll review the prototype API, show-case its features and highlight key differences with the existing one.
Read more here: pytorch.org/blog/introducing…
4
5