RSS Feeds

• France international says title ‘one of my biggest dreams’

• Club rejected weekend Gakpo bid from Manchester City

Bradley Barcola is aiming to win the Premier League after completing his move to Liverpool from Paris Saint‑Germain for an initial £106m, which could rise to £123m. The France international has signed a five-year deal.

Andoni Iraola has prioritised signing wingers this summer after replacing Arne Slot. Barcola follows the arrival of Victor Muñoz as the former Bournemouth head coach aims to rebuild the front line.

The performances often gave me goose bumps, but I also got to observe the highs and lows of audience behaviour

• Read more in this series here

I was 24, had just moved into a share house I couldn’t afford, and my job as a nanny was proving less than ideal. I’d graduated from my music degree a few years prior and was floating along, feeling a little lost.

A friend had told me to apply to become a front-of-house staff member at Australia’s busiest performing arts centre, Arts Centre Melbourne. It was a rare opportunity, as people rarely quit. To be surrounded by art again sounded like a dream.

A graphic novel by Melissa Chan and Badiucao imagines China and the US at war in 2035. They hope to remind young people that ‘authoritarianism everywhere is bad’

Drinking Tsingtao beers, eating congee, walking with hands behind their backs: for many young people in western countries, “Chinamaxxing” has become all the rage. Some content creators have even gone beyond the jokes about being in “a very Chinese time in my life” and have followed this passion to the Middle Kingdom: making TikToks in Chongqing’s multilayered metropolis, bathing in Chengdu’s 24-hour spas or backflipping on the Great Wall of China.

This irreverent, mostly ironic trend is jarring for some, but Chinese Australian artist Badiucao and Chinese American journalist Melissa Chan don’t want to lecture the so-called Chinamaxxers.

Fashion designer was convicted of hate crime by French court in 2011 after series of racist and antisemitic remarks

The controversial fashion designer John Galliano announced on Monday that an exhibition honoring his work that New York’s Metropolitan Museum of Art had scheduled to open for the upcoming spring’s Met Gala is not going to go forward.

The museum’s decision to honor Galliano, who was convicted of a hate crime by a French court in 2011 after a series of racist and antisemitic remarks, had drawn controversy and backlash.

Majority of missing Australians were near Gyirong Port minutes before impact. Rescuers say the area is now destroyed, covered in metres of silt and debris

• Follow our Australia news live blog for latest updates

• Get our breaking news email, free app or daily news podcast

It has been almost a week since the catastrophic floods in Nepal and not even a vehicle can get into the area where the majority of missing Australians were last heard from.

Gyirong Port, the border area between Nepal and China, was the first to be hit. Images from Chinese rescue teams who have managed to get in by helicopter or hiking show the area has been completely flattened. The immigration building, the power lines, everything, just wiped away.

These things are undermining not just our public spaces but the very nature of being human. Let’s shame them off the streets

So far I am yet to be tempted into purchasing a pair of Ray-Ban Meta glasses. To a certain extent this is a function of possessing a large, misshapen and comically convex head upon which no pair of sunglasses has ever looked remotely appropriate. Nor am I one of those consumers who professes to be “really into gadgets”, a preference I’ve always found weirdly unspecific. It’s subtly telling when people claim to be into “gadgets”, because they’re never talking about, say, a tin opener. What they really mean is something gimmicky and expensive that almost certainly requires the services of a datacentre and enough water to irrigate Kent.

And even if I were to surmount these two obstacles, you then get into the ethics of the thing. I’m nobody’s idea of a PR expert, but if your shiny new product keeps requiring patches and updates to prevent people from using it to do degenerate things to women, perhaps it might be worth walking it back to the research and development stage.

Jonathan Liew is a Guardian columnist

Fashion designer was convicted of hate crime by French court in 2011 after series of racist and antisemitic remarks

The controversial fashion designer John Galliano announced on Monday that an exhibition honoring his work that New York’s Metropolitan Museum of Art had scheduled to open for the upcoming spring’s Met Gala is not going to go forward.

The museum’s decision to honor Galliano, who was convicted of a hate crime by a French court in 2011 after a series of racist and antisemitic remarks, had drawn controversy and backlash.

Перевод средств в современном мире — дело пары минут: достаточно сообщить свой номер телефона — и деньги у вас. Но часто бывает так, что нам не хочется делиться номером с посторонними людьми. Например, когда нужно поучаствовать в сборе денег на подарки в школе или перевести деньги за выкупленный билет на концерт.

Для решения подобных задач мы в Мир Plat.From (НСПК) разработали функцию платёжной ссылки C2CQR, которая позволяет обойтись без номера телефона. В отличие от QR‑кода для платежей на кассах, этот QR предназначен для переводов между физлицами. Он генерируется в мобильном приложении банка и позволяет отправить деньги разными способами.

Это не просто новый шаг к упрощению платежей, а результат долгого развития технологий денежных переводов. Давайте вместе посмотрим, как люди отправляли друг другу деньги много лет назад, как мы облегчили процесс с помощью нового решения и что происходит под капотом, когда вместо номера телефона пользователь показывает платёжную ссылку C2C.

• Cape Town stunt sparks major debate over safety risks

• Airlink insists flyby conducted according to regulations

A pre-game stunt involving two passenger jets flying dramatically low over a stadium in South Africa drew huge cheers from the estimated crowd of 56,000 people, but also sparked a major debate over safety risks. The flyby by two Embraer jets, operated by the local airline Airlink, happened on Saturday before the second rugby union Test between South Africa and New Zealand in Cape Town.

The flight tracking service Flightradar24 estimated that the lead jet passed just 41-46 feet (12-14 metres) above the roof of Cape Town Stadium, with the crowd and the world’s top-ranked rugby teams below. The second jet, which was very close behind, flew through a cloud of pink dust that shot up from the stadium roof as part of the show.

Ситуация из практики любой компании: сотрудник уволился, ноутбук сдал, пароль от гостевой и рабочей сети остался прежним. Формально сотрудник его не знает — вводил один раз при подключении, полгода назад, наизусть не помнит.

На самом деле пароль он унёс: если ноутбук уехал с ним, а иногда и просто потому, что пароль можно было посмотреть в системе одной командой в любой момент.

Есть показатель, который получается, если совместить сразу два вопроса из опросов бизнеса Банка России: как компании планируют менять цены и как они планируют менять численность сотрудников. В проекте «Простая аналитика» считается разрыв между ними. Логика проста: чем он больше, тем сильнее бизнес готовится повышать цены, но при этом почти не собирается расширяться. И вот сейчас этот разрыв довольно большой.

По последним данным индекс Разрыв ожиданий: цены и занятость составляет 15,8 п.п. Сам показатель строится как разность двух сопоставимых балансов одного опроса предприятий Банка России и выходит ежеквартально, поэтому здесь важно смотреть именно на индекс, а не пытаться сопоставлять его с отдельным ежемесячным показателем ценовых ожиданий ЦБ. И 15,8 п.п. – это уже довольно заметный разрыв. Ожидания компаний по расширению численности сейчас находятся около нуля: то есть компании в среднем не планируют массово сокращать работников, но и прежнего желания активно расширять штат уже практически нет.

Внедрение OKR часто превращается в бюрократию. Сотрудники воспринимают его как тот же Jira, только в профиль, а рыночный софт либо стоит дорого и тащит за собой тонны ненужного HR-функционала, либо требует сборки на коленке.

В Sape мы столкнулись со всеми этими проблемами и решили, что OKR-софт должен быть простым, интуитивно понятным и не ломать текущие процессы в таск-трекерах. В результате мы написали свой open-source-инструмент, который собирает метрики через OpenAPI и отлично дружит с n8n.

Пятьдесят дней назад планирование моих тренировок переехало из головы и переписки в телеграм-бота. Он читает восстановление из WHOOP и форму из TrainingPeaks, рассуждает через Claude и сам пишет структурированные тренировки обратно в календарь — оттуда они уезжают в Garmin Connect и на часы. За это время он записал в календарь 72 тренировки, а стоимость эксплуатации (дроплет, инференс, подписки) составила около 4 500 рублей в месяц против 30 000+ за живого тренера с теми же подписками на спортивные сервисы.

Официального доступа к TrainingPeaks у меня нет: в партнёрскую программу я написал и получил отказ без объяснения причин. В статье — как я всё равно научился писать в чужой календарь: обмен куки веб-сессии на bearer, GET-merge-PUT вместо несуществующего PATCH, поле, которое на чтении приходит объектом, а на запись требует JSON внутри JSON, и почему IF и TSS плановой тренировки приходится считать самому.

Вторая половина — про то, где это ломалось. Три отказа за пятьдесят дней, и ни один не был виной модели: устаревший в настройках пороговый темп, мой собственный баг с зонами не того вида спорта, один невалидный OAuth-scope, отбивавший запрос согласия целиком, и ответы, обрывавшиеся на полуслове из-за параметра, у которого «по умолчанию» значит разное у разных моделей. Плюс три дня, когда я был уверен, что агент зря меня бережёт, — а данные говорят, что прав был он.

Спортивного результата пока нет: до целевого марафона сто дней, но надеюсь на лучшее.

Вы нажимаете в ERP кнопку «Рассчитать план», получаете плановые заказы с датами и резонно предполагаете, что система только что построила производственный план. Она его не строила. Она рассчитала, что нужно закупить и изготовить под план, который кто‑то дал ей на вход. Разбираю двенадцать характеристик производства, при которых MRP перестаёт справляться, и объясняю, откуда на самом деле должен браться производственный план.

Когда 180 тестов каждый раз проходят форму входа, авторизация начинает съедать десятки минут и ломаться на рейт‑лимитах, SSO и редиректах. Разберём рабочую схему: получить токен по API, закэшировать состояние и передать его браузеру до старта приложения.

Xabi Alonso’s side have scored seven and conceded five in two wins so far, a product of the squad’s constant churn

• Sign up for our free newsletter here

Two Premier League games played for Chelsea under Xabi Alonso: two wins and seven goals scored. It’s the sort of start that would delight many new managers – only two other clubs can boast about a six-point start this Premier League season (albeit with Arsenal to play later on Monday). Points gained in this odd transitional phase, when almost nobody seems quite ready, could prove extremely valuable come the end of the season. And yet, for all the attacking potency, there must also be concerns about Chelsea’s openness and the five goals conceded.

Context must also be considered, and it projects both ways. It’s early in the season. There has been, as ever, a lot of change at Chelsea. They will get better; they will achieve greater coherence. But equally there has to be a sense of realism about the nature of the teams they’ve played. Brighton may have thrashed a shambolic Aston Villa 4-0 last weekend, but they were missing seven frontline players at Stamford Bridge and still put three past Chelsea on Sunday. And while Fulham looked very lively as Chelsea won 3-2 on the opening weekend, they couldn’t muster a shot on target against Sunderland on Sunday.

This is an extract from Soccer Desk, a newsletter from the Guardian US. Subscribe for free here.

Ailing Djokovic goes out | Williams loses after 2am start
You can email Taha and follow us on TikTok and Bluesky

Sabalenka eases through on her serve, advancing to the net to secure her first game of the tournament with a forehand volley. It’s 1-1 in the opening set against Osorio.

Sabalenka and Osorio are underway on Arthur Ashe, with the Colombian serving. She records two double-faults before Sabalenka seizes upon the second serve, whipping a blistering forehand to make it deuce. But the defending champion finds the net and the stratosphere, too, as Osorio holds to go 1-0 up.

Model Context Protocol (MCP) от Anthropic перевернул работу с ИИ-агентами, но когда их становится больше трех, начинается ад с правами доступа, сетевым I/O и визуальным контролем. Мы задолбались собирать костыли на Python и написали масштабируемый опенсорс-оркестратор. Внутри — честный инженерный разбор: почему для ядра выбрали Go, как упаковали асинхронный движок JSON-RPC 2.0 на Rust в cdylib для вызова через cgo, и как мы героически побеждали утечки памяти на стыке этих двух миров.

In our previous blog post in this series, we discussed state-of-the-art models for object detection: the architectures, the theory, and what makes YOLO12, YOLO26, and RF-DETR tick. If you want the theoretical background on these models, start there.

This post is the practical follow-up: how to actually use these models, how to fine-tune them on diverse, specialized datasets that look nothing like their training data, and how to evaluate the results – all within PyCharm.

Why fine-tune at all?

Every pretrained detector you download was trained on some distribution of images, almost always COCO, which is ~118k training images of everyday scenes containing 80 common object categories (people, cars, dogs, chairs, etc.).

Real-world deployment data rarely looks like COCO. Things that object detection might actually be applied to, such as damaged industrial cables, bone fractures on X-rays, or densely stacked soda bottles on a shelf, are:

• Out of vocabulary: “Bone fracture” is not one of COCO’s 80 classes, so the model literally has no output category for it.
• Out of visual distribution: X-ray imagery, industrial close-ups, and heavily occluded shelf scenes differ drastically from consumer photos in texture, viewpoint, and object density.

Deploying a detector on off-distribution data therefore requires fine-tuning. But before we break the models, let’s establish that we get similar results on our hardware to the ones reported by developers.

The models

For the purposes of this experiment, we’ll focus on three current SOTA object detection families and examine two sizes of each model:

Sanity check: Reproducing COCO val2017 baselines

We’re going to be working with six pretrained checkpoints: Two different sizes of each of the three models. To check that these models are behaving as expected, we evaluated all of them on the full 5,000-image COCO validation dataset (val2017) to verify the numbers reported in the previous post:

Three things stand out even before we leave COCO behind:

1. Larger models (mostly) have better performance. RF-DETR Base leads (0.5325 mAP50-95), but the medium YOLOs get remarkably close (0.5259 / 0.5181) with ~10M fewer parameters.
2. YOLO26’s NMS-free design pays off in throughput. YOLO26-N is the fastest model in the lineup (12.3 ms latency) at essentially the same mAP50-95 as YOLOv12-N, which, despite being the smallest model here, is considerably slower (23.9 ms latency). Attention is expensive. (If you want more detail on the model architectures and how they affect performance, see the previous blog post in this series.)
3. RF-DETR Nano is not “nano” by parameter count (~30M – more than YOLO26-M), but it is well-optimized: With 12.4 ms latency, it is the second-fastest overall.

Published papers report optimized inference latency: That is, they measure the model’s forward pass in isolation, stripped of the surrounding stages of the object detection pipeline. We deliberately skipped that aggressive optimization so our numbers reflect what you’d actually see when deploying these models.

As a result, our latency figures don’t line up with the benchmarks in the models’ white papers. There are two main reasons for this:

• Hardware: We used different hardware from the NVIDIA T4 GPU that serves as the de facto standard in object detection benchmarking.
• Unoptimized computation graph: We ran the models in their native framework rather than converting them to TensorRT. TensorRT compiles a network into a hardware-specific engine, fusing layers, selecting the fastest kernels for the target GPU, and optionally running in reduced precision. That can cut latency substantially, but the resulting engine is tied to one GPU and requires an extra build step, so it doesn’t represent how these models perform out of the box.

Accuracy is a different story: While our latencies diverge from the published ones, our mAP50-95 results fall within reasonable noise bounds of the reported figures.

Now that we’ve seen what our pretrained models can do on COCO, the dataset they were trained on, let’s see what happens when they’re tested off distribution.

The datasets

For evaluation, we used RF100-VL, a large-scale collection of 100 multimodal datasets covering concepts deliberately chosen to be rare in object detection models’ pretraining data. These datasets contain exactly the off-distribution targets we care about. These targets also mirror common real-life applications for object detection, giving us a realistic test of these models’ capabilities out in the wild.

We picked three datasets that stress test the models in different ways:

Tutorial: Fine-tuning all three models in PyCharm 🙂

Step 1: Setting up the project

One of the first challenges we had to overcome in this project was that the three implementations do not share a compatible set of dependencies. In particular, the two different generations of YOLO require different versions of the ultralytics package. PyCharm offers a clean solution for this: one PyCharm project with three isolated uv environments – one per model family.

We’ll run our computations on a remote GPU. Configuring a remote interpreter in PyCharm follows the same workflow as a local one: the same dialog and the same dropdown as in the local case. Note that remote interpreters require PyCharm Professional; Community Edition supports local environments only.

Firstly, we need to instantiate our three uv virtual environments via:

cd yolov12 && uv venv .venv --python 3.11 

cd yolov26 && uv venv .venv --python 3.11 

cd rf-detr && uv venv .venv --python 3.11

Once your uv virtual environments exist, register each one as an existing interpreter. Go to Settings | Python | Interpreter, click Add Interpreter → Add Local Interpreter, choose Environment as Select existing, and point the interpreter field at that environment’s bin/python. PyCharm doesn’t create anything here, it just picks up the environment uv already built.

Repeat for each environment. From then on, switching is a matter of picking one from the Settings | Python | Interpreter dropdown, or from the interpreter widget in the bottom-right-hand status bar.

You can find the full list of dependencies required for each model in their respective project repositories. You can either install all the projects’ dependencies in PyCharm’s built-in Terminal tool window or install individual packages using the Python Packages tool window (including selecting specific versions of packages). You can access both of these tool windows by clicking the relevant icons in the lower left-hand corner of the PyCharm toolbar. 

For a step-by-step guide on setting up the environments for all three models, see our GitHub implementation of this tutorial.

Step 2: Getting the datasets

To obtain the out-of-COCO-distribution datasets, we can install our datasets via the rf-detr virtual environment, since it has roboflow as one of its core dependencies. We then set the Roboflow API key as an environment variable so that it is available to the API when downloading the datasets.

pip install roboflow

export ROBOFLOW_API_KEY="your_key_here" # you can get API key here: https://docs.roboflow.com/reference/authentication/authentication/find-your-roboflow-api-key 

After setting everything up, now you can run the Python script below to get the three datasets we’re going to use in our tutorial:

import os
from roboflow import Roboflow

api_key = os.environ.get("ROBOFLOW_API_KEY")

if not api_key:
    raise RuntimeError("ROBOFLOW_API_KEY is not set")

DATASETS = [
    "bone-fracture-7fylg",
    "cable-damage",
    "soda-bottles",
]

VERSION = 2          # RF100 projects are generally published at version 2
FORMAT = "yolov8"    # or "coco", "voc", "yolov5"
rf = Roboflow(api_key=api_key)
workspace = rf.workspace("rf100")

for slug in DATASETS:
    print(f"Downloading {slug} ...")

    try:
        project = workspace.project(slug)
        dataset = project.version(VERSION).download(FORMAT)
        print(f"  -> {dataset.location}")

    except Exception as e:
        print(f"  !! failed: {e}")

This script connects to the Roboflow cloud service via its Python API client and downloads three specified RF100 datasets in YOLOv8 format. It loops through each dataset, reports where successful downloads are saved, and prints an error if any download fails.

Step 3: Getting a zero-shot baseline by using pretrained models on custom data

Before fine-tuning, we’re going to evaluate the COCO-pretrained checkpoints directly on our three datasets, to see whether the fine-tuning is actually necessary. The result was unambiguous: The models predicted essentially nothing.

Zero-shot mAP50-95 on the test splits of our three datasets:

This is to be expected; it’s not a bug! As the models are closed-vocabulary detectors, that is, they have a finite number of predefined target classes, they physically cannot output a class like fracture that isn’t in their 80-class COCO head. 

This is the punchline of this whole post: A model scoring 0.72 mAP50 on COCO scores 0.00 on bone fractures. Pretrained ≠ deployable, even when the model is state of the art. Basic machine learning principles still apply, even in the age of AI!

Step 4: Fine-tuning

All models were fine-tuned on a single A100 GPU for 10 epochs. We used standard Ultralytics/RF-DETR fine-tuning pipelines in order to fine-tune the models on our three datasets. We fine-tuned a model for each dataset. The full fine-tuning pipeline can be found in finetune_rf100.py scripts in the project repo, under the folders for each model.

You can see the core of the training setup below. Both YOLO and RF-DETR are built on PyTorch under the hood, but the training loops are abstracted behind higher-level library APIs: Ultralytics’ YOLO.train() for the YOLO models, and RF-DETR’s own train() functionality.

YOLO12 and YOLO26

train_model = YOLO(args.model)

train_res = train_model.train(
                data=str(yaml_path),
                epochs=args.epochs,
                imgsz=args.imgsz,
                batch=args.batch,
                device=args.device,
                project=args.project,
                name=run_name,
                exist_ok=True,
                verbose=False,
            )

RF-DETR

ModelClass().train(
                dataset_dir=str(coco_dir),
                output_dir=str(output_dir),
                epochs=args.epochs,
                batch_size=args.batch_size,
                grad_accum_steps=args.grad_accum,
                lr=args.lr,
                resolution=resolution,
                early_stopping=True,
                checkpoint_interval=1,
            )

Step 5: Results

Fine-tuning transforms the picture. You can see the results on the test set after training:

On the left, we have the pretrained models’ results for the COCO validation dataset. As we showed earlier, accuracy (mAP50-95) fell between 0.39 and 0.53, and all models except for YOLOv12-M showed low latency. The fine-tuned models on the right showed a similar range of accuracy for the cable-damage and soda-bottle detection tasks, only falling lower for the bone-fracture task. Moreover, the fine-tuned models were comparable in latency to the pretrained models for their intended tasks, and for YOLOv12-M, they were even faster. This suggests that, after fine-tuning to the target domain, the models achieve performance that’s broadly comparable to the pretrained performance on their original training domain.

Let’s now have a closer look at the fine-tuned models’ performance, breaking it down by mAP50 and mAP50-95 for the three separate RF-100 datasets:

What the numbers say:

• The soda-bottles target is the easy win. Every model lands in the 0.91–0.97 mAP50 band. This is likely due to the fact that the domain (consumer products in photos) is visually close to existing classes in COCO, so only the vocabulary was new. Interestingly, the attention model family does great here, with YOLOv12-M taking the top spot (0.6422 mAP50-95).
• cable-damage: Detection is easy, but localization is hard. mAP50 reaches 0.93, but mAP50-95 tops out at 0.446. It appears that the models find the damage reliably, yet they struggle to box thin, elongated defects precisely. If your application needs tight boxes at high IoU, this gap would be a significant issue.
• bone-fracture remains genuinely hard. The best model (RF-DETR Base, 0.447 mAP50) is far from production-ready, and the performance spread across models is huge. The modality shift from photos to X-rays means the pretrained backbone features transfer poorly. The different image modality and small, sometimes almost indistinguishable bone fractures make the detection task way harder than the one employed on common objects identification. This is the dataset that would most benefit from domain-specific pretraining, more data, or longer fine-tuning. 
• RF-DETR Base is the most consistent performer, winning on two out of three datasets and challenging seriously for the third. The DETR-style architecture seems to transfer more robustly to unfamiliar domains. 

Qualitative results

To visually assess how these models perform, we can overlay the predicted bounding boxes on the images. Let’s look at the objects our models detected in six random images per class:

We can see this confirms the accuracy values we saw above: The noisy images of soda bottles in fridges are labeled accurately, with tight bounding boxes for each object. The cable damage is identified less consistently, with some models failing to find the damage altogether, and others creating unnecessarily large bounding boxes. Finally, the images of broken bones contrast sharply with the other two, with less than half of the images having any break identified, and different models identifying different potential breakage points.

Conclusions 

Pretrained object detectors are powerful, based on advancements in model architecture over the past five years, but as we’ve seen here, pretrained does not necessarily mean deployable. All six models performed well on COCO, yet when we applied those same checkpoints directly to our specialized datasets, their performance fell close to zero. However, fine-tuning completely changed that picture.

After only 10 epochs of fine-tuning, all three model families were able to adapt well to both the cable-damage and soda-bottle datasets. As we noted, the soda-bottle task was particularly transferable, likely because it contained objects similar to those contained in COCO. cable-damage was also detected relatively reliably, although the larger gap between mAP50 and mAP50-95 showed that precisely locating these tiny defects was still challenging for all of the models. However, bone-fracture was a completely different story, likely because moving from the sort of natural images contained in COCO to X-rays is a much larger domain shift. While RF-DETR handled this jump best, even its performance shows the limits of fine-tuning, and there are times when you might need to consider more data, longer training, or even domain-specific pretraining.

The broader takeaway is that there is no single “best” detector: It is dependent on the task. Model size, latency requirements, licensing restrictions, and most importantly, the similarity between the model’s pretraining data and your target domain all affect the outcome. It is important to refrain from unquestioningly trusting the numbers reported by model providers and explore the fit of a specific model for your own particular task.

Get started with PyCharm today

In this post, we’ve gone from validating pretrained YOLO12, YOLO26, and RF-DETR checkpoints on COCO to testing them zero-shot on specialized data, to fine-tuning them on three very different object detection tasks, and then finally, comparing the resulting accuracy and latency. Along the way, we’ve seen how PyCharm can help manage the practical side of a project like this, where multiple model families require different dependency sets and training environments.

PyCharm helps you keep these workflows together in a single project while using isolated Python environments for each model family. Its interpreter management, built-in terminal, Python Packages tool window, and support for remote development make it easier to move between environments and run training on remote GPU hardware without having to manage each part of this workflow separately.

If you’d like to try these experiments yourself, maybe look into fine-tuning these models for your own specific object detection use case! PyCharm is available to download and try. You can use the accompanying project code to reproduce our COCO baselines, download the RF100 datasets, fine-tune the models, and evaluate them using the held-out test splits.

You can find the full code for this project on GitHub. And if you’d like to learn more about object detection, including the architectures behind the models we used in this post, check out the previous post in this series.

Специалисты Лаборатории цифровой криминалистики и исследования вредоносного кода компании F6 предупреждают о новой угрозе российскому бизнесу. В августе 2026 года специалисты Лаборатории выявили активность группировки вымогателей, которая называет себя VantaCore. На данный момент известно как минимум о семи жертвах группировки. Суммы запрашиваемых выкупов составляют миллионы долларов.

Исследователи из канадского Университета Торонто на прошлой неделе опубликовали научную работу, в которой продемонстрировали новую атаку на видеоускорители NVIDIA. Атака, получившая название GPUThor, относится к классу Rowhammer, то есть использует многократные обращения к ячейкам оперативной памяти с целью повлиять на соседние ячейки. Таким образом можно изменить данные в областях памяти, изначально недоступных потенциальному злоумышленнику. Наиболее актуален такой сценарий атаки в случае совместного доступа к профессиональному видеоускорителю. Именно поэтому в подобных работах традиционно исследуются устройства NVIDIA, в данном случае модели поколения Ampere A4000, A4500, A5000 и A6000.

По сравнению с предыдущими атаками на подобные устройства, продемонстрированными в начале 2026 года, GPUThor обеспечивает изменение данных в целевых ячейках в сотни и даже тысячи раз чаще. Но самое главное — новая атака в некоторых случаях приводит к двойным и тройным бит-флипам, которые не корректируются системой ECC. Это точно позволяет провести атаку типа «отказ в обслуживании» и теоретически создает условия для атаки с повышением привилегий, даже если коррекция ошибок включена.

Merge queue гоняет полный сьют один раз на каждый пул-реквест в группе. Четверо ждут посадки — четыре полных прогона, причём последний из них уже содержит три остальных. Обязаны ли те три прогоняться? Мы построили одноразовый стенд, прогнали на нём четыре способа останавливать лишние прогоны и померили каждый: один сажает сломанный код, один залипает намертво, один безопасен и стоит ровно столько же, сколько мы платим сейчас. Четвёртый — наш, и он экономит около 19 машинных минут на посаженный пул-реквест ценой примерно четырёх минут ожидания.

People hit the streets of west London as Europe’s largest street carnival marks its 60th year, celebrating Caribbean history and culture