Daniel Moraes

Remote Sensing.
Earth Observation.
Machine Learning.

[Daniel Moraes @danielmoraesofc]

@itsaflecha Europa tem o projeto EuroLLM, Portugal tem o Amália, Espanha tem o Alia e o Salamandra. Mas o Brasil não pode, não sabe do que estão falando etc. Me poupe

[Daniel Moraes @danielmoraesofc]

@MaxLenormand Cool

[Daniel Moraes @danielmoraesofc]

@giswqs Looks great. Any tutorial on how to run it inside jupyter notebooks?

[Remote Sensing MDPI @RemoteSens_MDPI]

🌳🌳 A Weakly Supervised and Self-Supervised Learning Approach for Semantic #Segmentation of #LandCover in #Satellite Images with National #Forest Inventory #Data

✍️ Daniel Moraes et al.
🔗 https://t.co/fgblABkYsw

[Daniel Moraes @danielmoraesofc]

@P_droMenezes Salvador é a cidade mais relevante do Brasil culturalmente

[Daniel Moraes @danielmoraesofc]

@LMontenegro_PT @PauloRangel_pt Vergonhoso

[Daniel Moraes @danielmoraesofc]

@levelsio Private providers exist in the EU too. You're comparing two different things: private providers and EU-funded grants to foster AI adoption

[Daniel Moraes @danielmoraesofc]

@levelsio I'm pretty sure that's not the case for the AI Factories. Any European startup or SME should be eligible

[Daniel Moraes @danielmoraesofc]

@savage_agi @levelsio This. If you want to pay for GPUs go to a private provider. EU's AI Factory idea is to foster AI adoption by providing GPUs and user support. Of course that in order to be able to use it there should be some screening process. EU cannot simply give GPU hours for free

[Daniel Moraes @danielmoraesofc]

@levelsio The AI Factories will function more like an innovation and development ecosystem. What you are describing may fit more into the scope of the AI Gigafactories, which should be coming soon

[Daniel Moraes @danielmoraesofc]

@danielafcr_back Dicas de um seguidor soteropolitano: Casa de Jorge Amado (Rio Vermelho), Museu de Arte Moderna, Santo Antônio Além do Carmo

[Daniel Moraes @danielmoraesofc]

PhDone. I'm glad to have accomplished this goal and excited for the next chapters

[Daniel Moraes @danielmoraesofc]

Stop genocide
Free Palestine

[Daniel Moraes @danielmoraesofc]

📜Paper alert

We used Sentinel-2 & point-based NFI data as sparse labels for weakly supervised CNN-based national LC mapping. We also showed that pretraining a self-supervised MAE before finetuning with NFI data improved model performance.

🔗: https://t.co/2AHsbe8kTq

[Eric Berger @SciGuySpace]

If you care about science at NASA, this is royally sucks.

https://t.co/mjHH0dZVhG

[Daniel Moraes @danielmoraesofc]

@skdh They're not quite there yet. I was trying to use it to do a mini literature review and they do provide real references but they do not necessarily match with what the sentence before them says. Waste of time

[Daniel Moraes @danielmoraesofc]

📜Paper alert

We used Sentinel-2 & point-based NFI data as sparse labels for weakly supervised CNN-based national LC mapping. We also showed that pretraining a self-supervised MAE before finetuning with NFI data improved model performance.

🔗: https://t.co/2AHsbe8kTq

[Daniel Moraes @danielmoraesofc]

@gcamara @yohaniddawela But isn't naming categories a downstream task? A GFM could be trained on sensed data (spectral, elevation, temp, etc) only, with the semantic attribution being a downstream problem. Pretrained GFMs could provide informative features for fine-tuning, which is when semantics matter

[Daniel Moraes @danielmoraesofc]

Tate Modern

[Daniel Moraes @danielmoraesofc]

@jm_jonet Absurdo completo, vergonhoso