Field data collection for machine learning (ML) often relies on SD cards for storing images, documents, sensor logs, and other data. Unfortunately, it’s not uncommon to accidentally format an SD card, which wipes important training data. We will explain how to recover data from a formatted SD card, why recovery is often possible and how to prevent future data loss in ML workflows.
What Formatting Does and Can You Recover Files?
This means recovery software can often reconstruct files by scanning the physical card for recognizable data patterns (a process known as file carving).
However, if a full format or secure erase was performed, data recovery becomes impossible because these methods overwrite every sector with zeros. Most accidental formatting events, such as those occurring on drones or cameras, are quick formats, so recovery chances remain high.
How to Recover Formatted SD Card
Depending on how the card was formatted and what data was stored, you can often restore a large portion of your files, especially if you act quickly and avoid further use of the card. Below are three methods ML researchers and engineers can rely on to recover formatted SD card.
Method 1: Use Data Recovery Software to Recover a Formatted SD Card
There are a lot of different tools available. For this guide we will use Disk Drill. It is among the best SD card recovery software tools, suitable for restoring ML datasets. Disk Drill can restore images, videos, sensor logs, and structured data, because it supports various multiple file systems and formats (JPG, RAW, MP4, MOV, CSV, JSON). It also has Advanced Data Recovery mode, which is particularly useful for reconstructing fragmented videos.
Here how to recover data from formatted SD card using Disk Drill:
- Immediately remove the SD card from your device. Any further writes could overwrite the data you’re trying to restore.
- Visit the official CleverFiles Disk Drill website and download the version for your OS (Windows or macOS). Install the program by following the on-screen instructions.
- Insert the SD card into a card reader or adapter and connect it to your computer. Make sure it appears under “Drives” in Disk Drill’s main window.
- Select your SD card, click “Search for Lost Data,” and then choose the Universal Scan option – this is the most suitable mode for recovering data in the majority of cases, including formatted cards.
- Disk Drill will automatically start scanning for recoverable files. You can pause or preview files while the scan runs. Once the scan is complete, Disk Drill will organize results by file type (Pictures, Video, Audio, Documents, Archives).
- Preview files directly within the interface. Check the boxes beside the files or folders you want to restore and click Recover.
- Choose a different drive (not the SD card) as the recovery destination. Wait until the process completes.
Open sample files and make sure they’re readable. For partially corrupted videos, try rescanning the SD card using Advanced Camera Scan – this mode often brings back video fragments that don’t appear in regular scans.
If that doesn’t help, you can also try repairing the files with tools like FFmpeg, or extract usable frames to repurpose in your ML datasets. For CSV or JSON files, validate syntax and merge fragments with any backup copies.
Method 2: Recover Lost Files Using Backups
If you routinely back up your field data, this step is far simpler. Check all possible backup locations: external drives, lab servers, and cloud storage platforms such as Google Drive or AWS S3.
In team environments, one member’s workstation or synchronized cloud folder may already contain the missing files. Below is a step-by-step for Google Drive recovery:
- Log in to the account that normally receives uploads from the field device or backup script. This sounds obvious, but in multi-account setups (personal / lab / project shared drive), data often ends up in a different account than expected.
- Open the folder structure you normally use for data intake.
- Sort files by Last modified or Last opened to quickly surface recent uploads. Look for raw dumps like .jpg, .mp4, .csv, .json, .bin, .log, .zip, etc. These may not be in their final labeled dataset folder yet. Field engineers often drag and drop entire SD contents without renaming.
- If your team uses a Shared Drive (formerly Team Drive) or a project-wide shared folder, open that as well. Go to Shared with me and look for recent bulk uploads from teammates.
- Once you locate your backup copies, download them to a safe local drive (not back onto the formatted SD card).
Don’t forget to check deleted files, They stay in Trash for a limited retention window unless permanently removed.
Method 3: Consult a Professional Data Recovery Service
If software-based recovery fails or the SD card has physical damage (for example, bent connectors, cracked casing, or read errors), professional data recovery labs may be the only viable option.
These services use advanced imaging tools and chip-level access to extract data directly from the NAND memory. Professional recovery is expensive, but it’s often justified for irreplaceable ML datasets that cannot be re-collected. Before you send the card for service:
- Avoid further attempts with multiple tools, each scan can worsen physical wear.
- Clearly label the SD card and describe the circumstances of loss.
- Choose a certified recovery lab with experience in camera, drone, or IoT storage media.
Some labs also offer forensic-level imaging, which may be valuable for reconstructing partially corrupted scientific data or sensor recordings.
Preventive Strategies for Future Field Work
Data loss prevention is far easier and far less stressful than recovering files after an incident. In machine learning (ML) fieldwork, where datasets often represent unrepeatable moments or conditions, disciplined data management can make the difference between smooth progress and lost weeks of work.
Backup Frequently
Establish a clear backup schedule as part of every field operation. Data collected on SD cards should be copied at least once per day to a local device such as a laptop, portable SSD, or NAS.
In addition to physical storage, upload a copy to secure cloud storage whenever possible. You can use services like One Drive, Google Drive, Dropbox, or institutional servers. Ideally, maintain a 3-2-1 backup rule: three total copies of data, on two different types of media, with one copy stored offsite.
Use Redundant Recording
Whenever equipment allows, enable redundant recording. Many professional cameras, drones, and field loggers have two SD card slots that can mirror recordings in real time. This simple configuration provides immediate redundancy: if one card fails, the duplicate copy remains intact. In ML applications, this redundancy is especially valuable when collecting visual datasets.
Label and Track Cards
Disorganization often leads to accidental formatting or overwriting. Label each SD card with a unique identifier (e.g., ML01, ML02, etc.). Use waterproof stickers or laser marking. Maintain a card log sheet (digital or physical) to record when each card was issued, to whom, and when data was last offloaded.
For teams using multiple devices, color-coding by equipment type (drone, sensor unit) can further prevent confusion during busy field sessions.
Invest in Quality Media
Choose SD cards from reputable brands such as SanDisk or Lexar. High-quality cards offer faster read/write speeds and lower failure rates in extreme conditions. Avoid counterfeit or unbranded cards from online marketplaces.
Periodically test older cards and retire any that exhibit slow transfer speeds, file corruption, or repeated format prompts. The cost of a new card is negligible compared to the value of lost ML datasets.
Protect in Harsh Conditions
Environmental stress is a silent destroyer of field data. Exposure to moisture or heat can damage the SD card’s internal circuitry. Store cards in waterproof, shock-resistant cases and avoid direct sunlight or prolonged humidity. During expeditions in rugged terrain, keep cards in anti-static sleeves or weatherproof pouches. When removing cards from drones or sensors, power off devices first to prevent file system corruption caused by sudden disconnects.
Also, remember to design your data workflow with failure in mind. Create a disaster recovery plan that specifies what to do if a card is lost or corrupted: who to notify, which tools to use for recovery, and how to document recovered files. In long-term deployments, consider rotating backup media or sending verified copies back to a central repository at regular intervals.
Key Takeaways
Here are the key points to remember when dealing with a formatted SD card and planning data recovery:
- Quick formats rarely destroy data completely. They only reset the file system, leaving the actual data intact until new files overwrite it.
- Recovery is often possible if no new data has been written to the formatted SD card.
- Specialized recovery tools can scan the raw storage and rebuild files through deep analysis and file carving.
- Full formats or secure erases make recovery impossible because they overwrite every storage sector with new data.
- Preventive habits like regular backups and labeling SD cards reduce the risk of data loss in future field operations.