In the scenario of large-scale farming, accurate counting is a crucial part of production management, cost accounting, and risk prevention and control. However, traditional counting methods face numerous challenges:
(1) Dynamic Interference: Living animals exhibit complex behaviors such as occlusion, movement, and aggregation. For example, fish swim with overlapping bodies, and chickens gather together to feed. The frequent activities of living animals not only make manual counting difficult but also result in significant errors.
(2) Environmental Constraints: The aquaculture environment is complex, with constantly changing water turbidity. The lighting conditions in poultry houses fluctuate greatly, and poultry are easily disturbed by backgrounds with similar colors. Livestock in open pastures are not only vulnerable to background interference but also vary in size within a wide field of view. These factors not only cause visual fatigue to human eyes but also reduce the stability of traditional visual counting systems. Without being trained with data from the same environment, the counting accuracy drops sharply.
(3) Species Diversity: There are significant differences in body size among organisms at different growth stages and even among different species (for example, the body length of a bee is only 1.5 cm compared to that of an adult pig which exceeds 1 m). Traditional counting software requires individual training for each species, leading to a sharp increase in development costs.
These challenges are not only labor-consuming and resource-intensive for farmers and farming enterprises but also make it difficult to ensure the accuracy and timeliness of data, severely restricting the intelligent and refined development of the farming industry.
To address the counting challenges in the farming industry, CountAnything has joined hands with individual partners and enterprises to conduct iterative optimization in different application scenarios:
User Case1: Livestock/Poultry
The users of CountAnything include numerous individual ranchers and several livestock farming enterprises. They raise hundreds or even thousands of cattle and sheep on vast grasslands. Traditional manual counting not only consumes a great deal of manpower and time but also easily leads to missed counts or double counts in the complex terrain of the grassland where cattle and sheep are scattered. The counting software previously used by these users has poor stability in harsh environments such as strong light, backlight, and dust. Ranch workers are unable to keep track of the dynamic quantity of livestock in real time, which is extremely detrimental to the planning of breeding, fattening, and slaughtering.
"Even when using tools to take photos for counting, once we find any errors or omissions, by the time we make adjustments, the cattle and sheep will be in disarray, and we have to start all over again."
They used CountAnything to improve the entire process, reducing the counting time by 93%. The workers can now refocus their energy on more important matters such as livestock health and breeding, instead of wasting a lot of time and manpower dealing with the problem of livestock counting.
Figure 1: Livestock Counting on the Ranch
Similarly, for poultry counting, CountAnything can effectively mitigate the special lighting conditions and background interference in poultry houses, helping farmers to count poultry more accurately by taking photos:
Figure 2: Poultry Counting
User Case2: Aquatic Product Trading
CountAnything has a group of loyal users, including aquaculture farmers and wholesalers. The counting environment for aquatic products is complex. The reflection on the water surface, the dense overlapping of living organisms (such as fish and shrimp), and the turbid water quality all pose difficulties for accurate counting. In the sales process, due to the inability to count accurately, there are often disputes between farmers and wholesalers over the quantity, affecting the transaction efficiency and the relationship between the two parties.
Thanks to the powerful capabilities of the T-Rex2 model and the optimization for the aquatic environment, CountAnything can accurately identify and count even in the case of water surface reflection and dense overlapping of organisms.
"Before selling shrimp, we could only grab a netful and estimate the quantity. The buyers always suspected that the number of our shrimp was insufficient. If we counted them one by one, there were too many shrimp to count, and the more we counted, the more likely the shrimp were to die."
Now, shrimp farmers only need to take photos with their mobile phones, and CountAnything can provide an accurate quantity within a few seconds. This not only saves a large amount of manpower and reduces the mortality rate of aquatic products during the processing but also improves the transparency and efficiency of transactions. Buyers can verify the quantity through the software, eliminating the trust issues between the two parties.
Figure 3: Live Shrimp Counting
CountAnything has collaborated with small and medium-sized farmers and enterprises. By delving into the farming scenarios and listening to user feedback, it has provided efficient and accurate counting solutions for many farmers. In addition, by integrating CountAnything into the counting workflow, even small teams and individuals can enjoy the benefits and advantages of intelligence and automation that large enterprises have, at a lower cost. This enables refined farming, achieving cost reduction and efficiency improvement.
References
(1) Paper: "T-Rex2: Towards Generic Object Detection via Text-Visual Prompt Synergy" by Qing Jiang, Feng Li, Zhaoyang Zeng, Tianhe Ren, Shilong Liu, Lei Zhang. Link: https://arxiv.org/abs/2403.14610
(2) Access T-Rex2 API through DINO-X Platform: https://cloud.deepdataspace.com/
(3) CountAnything, the fast & precise counting tool based on T-Rex2: https://deepdataspace.com/products/countanything